Pith. sign in

REVIEW 3 major objections 3 minor 43 references

A compact model, TVF-DiT, hits 86.6% average success across three lab-manipulation tasks.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 19:43 UTC pith:VZZV344N

load-bearing objection Potentially useful compact imitation learning pipeline, but the central 86.6% claim is unverifiable from the abstract alone; worth a peer-review look if the full paper shows a fair evaluation. the 3 major comments →

arxiv 2603.01110 v2 pith:VZZV344N submitted 2026-03-01 cs.RO

Compact Task-Aligned Imitation Learning for Laboratory Automation

Kanata Suzuki , Hanon Nakamura , Kana Miyamoto , Tetsuya Ogata This is my paper
classification cs.RO
keywords imitation learninglaboratory automationrobot manipulationvision foundation modelvision-language modeldiffusion transformercompact modelprompt conditioning
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper sets out to show that laboratory automation does not require large, compute-hungry models. It introduces TVF-DiT, a framework under 500 million parameters that aligns a self-supervised vision foundation model with a vision-language model through a compact adapter and feeds the aligned representation into a Diffusion Transformer action expert. On three real bench tasks — test tube cleaning, test tube arrangement, and powder transfer — the authors report an average success rate of 86.6%, above alternative lightweight baselines. They also find that more detailed task prompts improve alignment and performance. A sympathetic reading: small foundation models, properly coupled, are enough for useful lab robots in low-VRAM settings.

Core claim

The central claim is that proper alignment, not raw model scale, makes imitation learning work for laboratory automation. TVF-DiT keeps the total parameter count under 500 million by freezing a self-supervised vision backbone and a vision-language model, connecting them with a compact adapter, and using a Diffusion Transformer as the action decoder. The reported result is an average 86.6% success rate across three physical tasks, with detailed prompts measurably boosting performance. The intended conclusion is that this recipe transfers across small manipulator tasks and is practical on GPUs with limited VRAM.

What carries the argument

TVF-DiT's load-bearing piece is the compact adapter that aligns the visual tokens of a self-supervised vision foundation model with the embedding space of a vision-language model. The aligned cross-modal representation is then used to condition a Diffusion Transformer action expert. This lets a small language-guided vision model steer action generation without training a large multi-modal fusion module, which is what keeps the whole system under 500 million parameters and therefore low-VRAM deployable.

Load-bearing premise

The load-bearing premise is that the evaluation — the three chosen tasks, the unnamed lightweight baselines, and the undisclosed success criteria — is representative and fair enough for the reported 86.6% average to support the conclusion that compact models suffice for practical laboratory automation.

What would settle it

A direct replication that names and re-runs the baselines under identical success criteria, and reports per-task trial counts and variance, would confirm or refute the comparative claim. A simpler check: degrade the prompts to generic one-word labels; if success does not drop appreciably, the paper's prompt-alignment mechanism is not doing the claimed work.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Laboratory robots can be trained from demonstrations with a sub-500M-parameter policy, requiring only low-VRAM GPU inference.
  • Prompt wording and detail become a direct performance lever: more detailed task prompts improve vision-language alignment and task success.
  • The same compact alignment recipe is claimed to generalize to other lab manipulation tasks beyond the three evaluated.
  • Small foundation models, aligned through adapters and paired with diffusion policies, offer a plausible route to cheaper lab automation without specialized hardware interfaces.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: if prompt detail is causal, prompt engineering could be treated as a tunable hyperparameter, and systematically varying prompt specificity on a held-out task would reveal whether the effect scales.
  • Beyond the paper: the 86.6% figure is a point estimate; with trial counts and per-task variance absent, a fair comparison to named baselines is the next necessary check before generalizing the claim.
  • Beyond the paper: the adapter-alignment design suggests a broader recipe — frozen perception and language models with a small learned bridge — that might transfer to other resource-limited embodied domains such as surgical or field robotics.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 3 minor

Summary. The manuscript proposes TVF-DiT, a sub-500M-parameter imitation learning pipeline for laboratory automation that aligns a self-supervised vision foundation model with a vision-language model via a compact adapter and integrates a Diffusion Transformer action expert. The abstract reports an average success rate of 86.6% on three real-world tasks (test tube cleaning, arrangement, and powder transfer), claims significant improvement over unspecified lightweight baselines, and states that detailed task prompts improve performance.

Significance. If the empirical claims hold, the work would demonstrate that compact foundation-model pipelines can support practical laboratory automation on limited computational resources, which is a useful contribution to low-cost robotics. However, the submitted manuscript contains no full text, methods, or experimental details. The claimed advance is entirely empirical, and none of the evidence needed to evaluate it is present. The idea is plausible and potentially valuable, but its significance cannot be assessed from the current submission.

major comments (3)
  1. [Full text (missing)] The manuscript as provided consists only of an abstract. There is no main text: no method description, no implementation details, no experimental protocol, no results tables, and no discussion. The central claims are therefore not verifiable. This is a load-bearing omission because the paper's contribution is an empirical demonstration; the abstract alone cannot support it. The authors should provide the complete manuscript with full experimental details.
  2. [Abstract, results sentence] The headline claim — 'an average success rate of 86.6%, significantly outperforming alternative lightweight baselines' — is reported without any supporting statistics: no trial counts, per-task success rates, confidence intervals, standard deviations, or success criteria. The names and configurations of the 'lightweight baselines' are also omitted. Without these, the claim is unfalsifiable: the result could reflect lenient scoring, favorable task selection, or weak baselines. At minimum, the authors must specify the number of trials per task, the success metric, and the baseline models.
  3. [Abstract, prompt-detail claim] The statement that 'detailed task prompts improve vision-language alignment and task performance' appears without a controlled comparison. It is unclear whether prompt detail was varied systematically, how 'detail' was quantified, and whether the effect is statistically significant. This claim is a secondary contribution but still requires experimental support. The authors should report the prompt conditions, the number of runs, and the effect size.
minor comments (3)
  1. [Abstract] The acronym TVF-DiT is not expanded; the full name of the framework would help readers.
  2. [Abstract] The phrase 'low-VRAM GPUs' is vague; specify the GPU model and memory footprint used for inference.
  3. [Abstract] The three tasks are named but not described; include task difficulty or success criteria in the full text, and reference the relevant figures or tables.

Circularity Check

0 steps flagged

No circular dependency found; the abstract reports empirical benchmark results rather than a derivation that re-uses its own outputs.

full rationale

The available text (the abstract) contains no derivation chain, fitted parameters renamed as predictions, or self-citations. The proposed method TVF-DiT is described as aligning a self-supervised vision foundation model with a vision-language model via a compact adapter and integrating them with a Diffusion Transformer policy. The headline result is an average task success rate of 86.6% across three real-world laboratory tasks. Success is measured by downstream task completion, which is an external metric, not a quantity re-inserted into the model's training objective. The claim that detailed task prompts improve performance is also an empirical observation, and no evidence is presented that prompts were tuned against the evaluation tasks; even if they were, that would be an experimental-design concern (possible overfitting of prompts), not a circular derivation. The lack of trial counts, variance, and baseline identities is a serious correctness/falsifiability issue, but it does not make the argument circular. No self-citation is used to justify the central claim, and no equation is shown to reduce to an input. Therefore, under the hard rules requiring specific quoted reduction or fitted-parameter renaming, the appropriate finding is no significant circularity.

Axiom & Free-Parameter Ledger

2 free parameters · 3 axioms · 0 invented entities

Abstract-only review. No new physical entities are proposed; the compact adapter is a design component, not an invented entity. The listed premises are the load-bearing assumptions visible from the abstract: demonstration data adequacy, baseline fairness, and the generalization from three tasks to the whole domain. Architecture hyperparameters and prompt-detail choices are the visible tuning knobs; their values and selection rules are unreported.

free parameters (2)
  • task prompt detail level = not reported
    The abstract reports that more detailed task prompts improve vision-language alignment and task performance; the prompt detail is a user-chosen input that materially changes results, with no stated principle for setting it, and it may have been tuned against the evaluation tasks.
  • model architecture hyperparameters = total under 500M parameters
    The advertised constraint is a sub-500M total budget, but the adapter width, transformer depth, training schedule, and alignment-loss weights are not described in the abstract; how these were chosen is invisible to the reviewer, and they are by-hand design choices.
axioms (3)
  • domain assumption Demonstration data for the three laboratory tasks adequately covers each task for imitation learning
    Imitation learning from demonstrations is the core method; the abstract gives no information about demonstration quantity, quality, or collection protocol, so the method's success depends on this unstated premise.
  • domain assumption The 'alternative lightweight baselines' are representative and fairly compared
    The headline claim ('significantly outperforming alternative lightweight baselines') depends on baseline selection and evaluation fairness; neither is described in the abstract.
  • domain assumption Success on three real-world tasks generalizes to 'practical laboratory automation'
    The final implication sentence extends from test tube cleaning, test tube arrangement, and powder transfer to laboratory automation broadly, an extrapolation that the three-task evidence alone does not establish.

pith-pipeline@v1.3.0-alltime-deepseek · 485 in / 14950 out tokens · 132422 ms · 2026-08-02T19:43:10.751868+00:00 · methodology

0 comments
read the original abstract

Robotic laboratory automation has traditionally relied on carefully engineered motion pipelines and task-specific hardware interfaces, resulting in high design cost and limited flexibility. While recent imitation learning techniques can generate general robot behaviors, their large model sizes often require high-performance computational resources, limiting applicability in practical laboratory environments. In this study, we propose a compact imitation learning framework for laboratory automation using small foundation models. The proposed method, TVF-DiT, aligns a self-supervised vision foundation model with a vision-language model through a compact adapter, and integrates them with a Diffusion Transformer-based action expert. The entire model consists of fewer than 500M parameters, enabling inference on low-VRAM GPUs. Experiments on three real-world laboratory tasks - test tube cleaning, test tube arrangement, and powder transfer - demonstrate an average success rate of 86.6%, significantly outperforming alternative lightweight baselines. Furthermore, detailed task prompts improve vision-language alignment and task performance. These results indicate that small foundation models, when properly aligned and integrated with diffusion-based policy learning, can effectively support practical laboratory automation with limited computational resources.

Figures

Figures reproduced from arXiv: 2603.01110 by Hanon Nakamura, Kana Miyamoto, Kanata Suzuki, Tetsuya Ogata.

Figure 1
Figure 1. Figure 1: Overview of this study for a laboratory automation framework [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Architecture of the proposed TVF-DiT framework. DINOv3 and SigLIP2 extract geometric and language-aligned representations, which are fused [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Task 1: Test tube cleaning requiring precise insertion and [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Representative execution sequences generated by the proposed method. Successful trials demonstrate continuous and coordinated manipulation [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Joint trajectories and 3D end-effector (EE) paths during inference. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Recovery behavior under external disturbance during test tube cleaning. When the tube orientation is manually perturbed, the model re-generates [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

43 extracted references · 14 linked inside Pith

  1. [1]

    11em plus .33em minus .07em @technote 4000 4000 100 4000 4000 500 `\.=1000 = #1 #1 #1 0pt [0pt][0pt] #1 * \| ** #1 \@IEEEauthorblockNstyle \@IEEEauthorblockAstyle \@IEEEauthordefaulttextstyle \@IEEEauthorblockconfadjspace -0.25em \@IEEEauthorblockNtopspace 0.0ex \@IEEEauthorblockAtopspace 0.0ex \@IEEEauthorblockNinterlinespace 2.6ex \@IEEEauthorblockAinte...

  2. [2]

    R. D. King, J. Rowland, S. G. Oliver, M. Young, W. Aubrey, E. Byrne, M. Liakata, M. Markham, P. Pir, L. N. Soldatova et al., ``The automation of science,'' Science, vol. 324, no. 5923, pp. 85--89, 2009

    2009

  3. [3]

    Steiner, J

    S. Steiner, J. Wolf, S. Glatzel, A. Andreou, J. M. Granda, G. Keenan, T. Hinkley, G. Aragon-Camarasa, P. J. Kitson, D. Angelone et al., ``Organic synthesis in a modular robotic system driven by a chemical programming language,'' Science, vol. 363, no. 6423, p. eaav2211, 2019

    2019

  4. [4]

    Burger, P

    B. Burger, P. M. Maffettone, V. V. Gusev, C. M. Aitchison, Y. Bai, X. Wang, X. Li, B. M. Alston, B. Li, R. Clowes et al., ``A mobile robotic chemist,'' Nature, vol. 583, no. 7815, pp. 237--241, 2020

    2020

  5. [5]

    N. J. Szymanski, B. Rendy, Y. Fei, R. E. Kumar, T. He, D. Milsted, M. J. McDermott, M. Gallant, E. D. Cubuk, A. Merchant et al., ``An autonomous laboratory for the accelerated synthesis of novel materials,'' Nature, vol. 624, no. 7990, pp. 86--91, 2023

    2023

  6. [6]

    B. P. MacLeod, F. G. Parlane, T. D. Morrissey, F. H \"a se, L. M. Roch, K. E. Dettelbach, R. Moreira, L. P. Yunker, M. B. Rooney, J. R. Deeth et al., ``Self-driving laboratory for accelerated discovery of thin-film materials,'' Science Advances, vol. 6, no. 20, p. eaaz8867, 2020

    2020

  7. [7]

    J. M. Granda, L. Donina, V. Dragone, D.-L. Long, and L. Cronin, ``Controlling an organic synthesis robot with machine learning to search for new reactivity,'' Nature, vol. 559, no. 7714, pp. 377--381, 2018

    2018

  8. [8]

    C. W. Coley, N. S. Eyke, and K. F. Jensen, ``Autonomous discovery in the chemical sciences part i: Progress,'' Angewandte Chemie International Edition, vol. 59, no. 51, pp. 22\,858--22\,893, 2020

    2020

  9. [9]

    G. N. Kanda, T. Tsuzuki, M. Terada, N. Sakai, N. Motozawa, T. Masuda, M. Nishida, C. T. Watanabe, T. Higashi, S. A. Horiguchi et al., ``Robotic search for optimal cell culture in regenerative medicine,'' Elife, vol. 11, p. e77007, 2022

    2022

  10. [10]

    F. Hase, L. M. Roch, C. Kreisbeck, and A. Aspuru-Guzik, ``Phoenics: a bayesian optimizer for chemistry,'' ACS central science, vol. 4, no. 9, pp. 1134--1145, 2018

    2018

  11. [11]

    Suzuki, H

    K. Suzuki, H. Ito, T. Yamada, K. Kase, and T. Ogata, ``Deep predictive learning: Motion learning concept inspired by cognitive robotics,'' arXiv preprint arXiv:2306.14714, 2023

    Pith/arXiv arXiv 2023

  12. [12]

    M. J. Kim, K. Pertsch, S. Oh, S. Singh, S. Nasiriany, D. Shah, V. Kumar, A. Xie, S. Levine, C. Finn et al., ``Openvla: An open-source vision-language-action model,'' arXiv preprint arXiv:2406.09246, 2024

    Pith/arXiv arXiv 2024

  13. [13]

    O. X.-E. Collaboration, A. O'Neill, A. Rehman, 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,'' in Proceedings of the 2024 IEEE International Conference on Robotics and Automation, 2024

    2024

  14. [14]

    Physical Intelligence et al., `` _ 0.5 : a vision-language-action model with open-world generalization,'' arXiv preprint arXiv:2504.16054, 2025

    Pith/arXiv arXiv 2025

  15. [15]

    Grattafiori, A

    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

    Pith/arXiv arXiv 2024

  16. [16]

    C. Wu, J. Li, J. Zhou, J. Lin, K. Gao, K. Yan, S.-m. Yin, S. Bai, X. Xu, Y. Chen et al., ``Qwen-image technical report,'' arXiv preprint arXiv:2508.02324, 2025

    Pith/arXiv arXiv 2025

  17. [17]

    Beyer, A

    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

    Pith/arXiv arXiv 2024

  18. [18]

    Sasaki, N

    H. Sasaki, N. Komeno, T. Hachimine, K. Takahashi, Y.-y. Ohnishi, T. Sugawara, A. Wakiuchi, M. Hatanaka, T. Miyao, H. Ajiro et al., ``Robotic system for chemical experiment automation with dual demonstration of end-effector and jig operations,'' International Journal of Intelligent Robotics and Applications, pp. 1--23, 2025

    2025

  19. [19]

    L. M. Roch, F. H \"a se, C. Kreisbeck, T. Tamayo-Mendoza, L. P. Yunker, J. E. Hein, and A. Aspuru-Guzik, ``Chemos: orchestrating autonomous experimentation,'' Science Robotics, vol. 3, no. 19, p. eaat5559, 2018

    2018

  20. [20]

    Christensen, L

    M. Christensen, L. P. Yunker, F. Adedeji, F. H \"a se, L. M. Roch, T. Gensch, G. dos Passos Gomes, T. Zepel, M. S. Sigman, A. Aspuru-Guzik et al., ``Data-science driven autonomous process optimization,'' Communications Chemistry, vol. 4, no. 1, p. 112, 2021

    2021

  21. [21]

    Pizzuto, H

    G. Pizzuto, H. Wang, H. Fakhruldeen, B. Peng, K. S. Luck, and A. I. Cooper, ``Accelerating laboratory automation through robot skill learning for sample scraping,'' in Proceedings of the 2024 IEEE 20th International Conference on Automation Science and Engineering, 2024, pp. 2103--2110

    2024

  22. [22]

    Kadokawa, M

    Y. Kadokawa, M. Hamaya, and K. Tanaka, ``Learning robotic powder weighing from simulation for laboratory automation,'' in Proceedings of the 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2023, pp. 2932--2939

    2023

  23. [23]

    Yamaguchi, C

    A. Yamaguchi, C. G. Atkeson, S. Niekum, and T. Ogasawara, ``Learning pouring skills from demonstration and practice,'' in Proceedings of the 2014 IEEE-RAS International Conference on Humanoid Robots, 2014, pp. 908--915

    2014

  24. [24]

    S. Wang, X. Cheng, J. Shiomi, and Y. Asano, ``Pipetting automation with remote pipette manipulation and learning-based tip recognition,'' in The Proceedings of JSME annual Conference on Robotics and Mechatronics, 2025, pp. 1P1--C10

    2025

  25. [25]

    Schenck and D

    C. Schenck and D. Fox, ``Reasoning about liquids via closed-loop simulation,'' arXiv preprint arXiv:1703.01656, 2017

    Pith/arXiv arXiv 2017

  26. [26]

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

    2024

  27. [27]

    T. Zhao, V. Kumar, S. Levine, and C. Finn, ``Learning fine-grained bimanual manipulation with low-cost hardware,'' in Proceedings of the Robotics: Science and Systems, 2023

    2023

  28. [28]

    Toyoda, K

    M. Toyoda, K. Suzuki, Y. Hayashi, and T. Ogata, ``Learning bidirectional translation between descriptions and actions with small paired data,'' IEEE Robotics and Automation Letters, vol. 7, no. 4, pp. 10\,930--10\,937, 2022

    2022

  29. [29]

    Suzuki and T

    K. Suzuki and T. Ogata, ``Sensorimotor attention and language-based regressions in shared latent variables for integrating robot motion learning and llm,'' in Proceedings of the 2024 IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, 2024

    2024

  30. [30]

    X. Liu, Z. Bai, H. Ci, K. Y. Ma, and M. Z. Shou, ``World-vla-loop: Closed-loop learning of video world model and vla policy,'' arXiv preprint arXiv:2602.06508, 2026

    Pith/arXiv arXiv 2026

  31. [31]

    M. J. Kim, Y. Gao, T.-Y. Lin, Y.-C. Lin, Y. Ge, G. Lam, P. Liang, S. Song, M.-Y. Liu, C. Finn et al., ``Cosmos policy: Fine-tuning video models for visuomotor control and planning,'' arXiv preprint arXiv:2601.16163, 2026

    Pith/arXiv arXiv 2026

  32. [32]

    Y. Xu, Y. Yang, Z. Fan, Y. Liu, Y. Li, B. Li, and Z. Zhang, ``Qvla: Not all channels are equal in vision-language-action model's quantization,'' arXiv preprint arXiv:2602.03782, 2026

    arXiv 2026

  33. [33]

    Sim \'e oni, H

    O. Sim \'e oni, H. V. Vo, M. Seitzer, F. Baldassarre, M. Oquab, C. Jose, V. Khalidov, M. Szafraniec, S. Yi, M. Ramamonjisoa et al., ``Dinov3,'' arXiv preprint arXiv:2508.10104, 2025

    Pith/arXiv arXiv 2025

  34. [34]

    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, 2023, pp. 11\,975--11\,986

    2023

  35. [35]

    Z. Qiu, Z. Huang, K. Wen, P. Jin, B. Zheng, Y. Zhou, H. Huang, Z. Wang, X. Li, H. Zhang et al., ``A unified view of attention and residual sinks: Outlier-driven rescaling is essential for transformer training,'' arXiv preprint arXiv:2601.22966, 2026

    arXiv 2026

  36. [36]

    Peebles and S

    W. Peebles and S. Xie, ``Scalable diffusion models with transformers,'' in Proceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 4195--4205

    2023

  37. [37]

    Lipman, R

    Y. Lipman, R. T. Chen, H. Ben-Hamu, M. Nickel, and M. Le, ``Flow matching for generative modeling,'' arXiv preprint arXiv:2210.02747, 2022

    Pith/arXiv arXiv 2022

  38. [38]

    Liu, ``Rectified flow: A marginal preserving approach to optimal transport,'' arXiv preprint arXiv:2209.14577, 2022

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

    Pith/arXiv arXiv 2022

  39. [39]

    AgileX, ``Cobot magic,'' Accessed:2026-02-21, https://global.agilex.ai/products/cobot-magic

    2026

  40. [40]

    Fujii, K

    W. Fujii, K. Suzuki, T. Ando, A. Tateishi, H. Mori, and T. Ogata, ``Buttoning task with a dual-arm robot: An exploratory study on a marker-based algorithmic method and marker-less machine learning methods,'' in Proceedings of the 2022 IEEE/SICE International Symposium on System Integration, 2022, pp. 682--689

    2022

  41. [41]

    Saito, M

    N. Saito, M. Tatsumi, A. Kubo, K. Suzuki, H. Ito, S. Sugano, and T. Ogata, ``Learning multimodal attention for manipulating deformable objects with changing states,'' in Proceedings of the 2025 IEEE-RAS 24th International Conference on Humanoid Robots, 2025, pp. 460--467

    2025

  42. [42]

    Marafioti, O

    A. Marafioti, O. Zohar, M. Farr \'e , M. Noyan, E. Bakouch, P. Cuenca, C. Zakka, L. B. Allal, A. Lozhkov, N. Tazi et al., ``Smolvlm: Redefining small and efficient multimodal models,'' arXiv preprint arXiv:2504.05299, 2025

    Pith/arXiv arXiv 2025

  43. [43]

    L. B. Allal, A. Lozhkov, E. Bakouch, G. M. Bl \'a zquez, G. Penedo, L. Tunstall, A. Marafioti, H. Kydl \' c ek, A. P. Lajar \' n, V. Srivastav et al., ``Smollm2: When smol goes big--data-centric training of a small language model,'' arXiv preprint arXiv:2502.02737, 2025

    Pith/arXiv arXiv 2025