Pith. sign in

REVIEW 3 major objections 3 minor 9 cited by

GraphCoT-VLA combines structured chain-of-thought reasoning with a real-time 3D Pose-Object graph so robot vision-language-action models can handle ambiguous instructions and unfamiliar environments.

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 ↗

GraphCoT-VLA couples structured chain-of-thought planning and a real-time 3D object-pose graph to improve robot manipulation under vague instructions.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection Abstract promises a novel VLA architecture that pairs structured CoT with a real-time 3D pose graph, but no evidence is given; the full paper could be good or could be another overclaimed robot model. the 3 major comments →

arxiv 2508.07650 v2 pith:WC7ENKS4 submitted 2025-08-11 cs.RO

GraphCoT-VLA: A 3D Spatial-Aware Reasoning Vision-Language-Action Model for Robotic Manipulation with Ambiguous Instructions

Helong Huang , Min Cen , Kai Tan , Xingyue Quan , Guowei Huang , Hong Zhang This is my paper
classification cs.RO
keywords vision-language-action model3D Pose-Object graphchain-of-thought reasoningrobotic manipulationambiguous instructionsreal-time spatial reasoningdropout hybrid reasoningopen-world generalization
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.

The reading

The paper argues that existing vision-language-action models struggle with ambiguous language and unknown environmental states because their perception is mostly static and two-dimensional. To fix this, it proposes GraphCoT-VLA, an end-to-end model that adds a structured Chain-of-Thought reasoning module plus a real-time updatable 3D Pose-Object graph capturing robot joints and object topology. On real-world robotic tasks, the paper reports that GraphCoT-VLA significantly outperforms existing methods in task success rate and response speed, with strong generalization under uncertain instructions. A sympathetic reader would see the central bet as: giving the model an explicit, live 3D spatial structure makes vague instructions resolvable and manipulation more robust.

Core claim

The paper's central claim is that an end-to-end vision-language-action model can overcome ambiguous instructions and unknown environments by combining two components. First, a structured Chain-of-Thought reasoning module integrates high-level task understanding and planning, feedback from failed tasks, and low-level imaginative reasoning about future object positions and robot actions. Second, a real-time updatable 3D Pose-Object graph represents the spatial configuration of robot joints and the topological relationships between objects, letting the model reason about 3D interactions rather than static 2D observations. A dropout hybrid reasoning strategy then produces efficient control outpu

What carries the argument

The central mechanism is the real-time updatable 3D Pose-Object graph: a live spatial representation of robot joint configurations and the topological relationships between objects, which carries the burden of grounding manipulation in 3D space. Around it, the structured Chain-of-Thought module supplies the reasoning chain from instruction to plan and to imagined future object positions and actions, while the dropout hybrid reasoning strategy trades reasoning depth against control latency. Together they let the model interpret ambiguous instructions and update its spatial understanding as the environment changes.

Load-bearing premise

The claimed gains depend on building and continuously updating the 3D Pose-Object graph from the robot's own observations with enough pose and topology accuracy that errors do not accumulate during a task.

What would settle it

Run a variant of the method with the 3D Pose-Object graph frozen at the first timestep: if task success on ambiguous instructions does not drop materially, the real-time graph is not doing the claimed work. Also, log pose and topology error over long horizons; if error grows with task length and correlates with failures, the real-time update claim fails.

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

If this is right

  • A single end-to-end VLA model can fold explicit 3D spatial structure and textual reasoning into control outputs, rather than treating perception and planning as separate static stages.
  • Ambiguous instructions can be resolved by explicitly generating possible future object positions and robot actions, giving the model a concrete hypothesis to test against the live 3D graph.
  • A real-time updatable 3D Pose-Object graph offers a path to handling open environments without per-scene retraining, since the model can revise object topology as observations change.
  • The dropout hybrid reasoning strategy makes chain-of-thought style reasoning fast enough for real-time robot control while retaining its planning benefit.
  • If replicated, the reported gains would push VLA research toward explicit 3D world models rather than reliance on static 2D visual features.

Where Pith is reading between the lines

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

  • A natural extension the paper leaves implicit is to measure how the 3D graph's pose errors accumulate over long horizons; the real-time update claim would be strengthened by showing that graph drift stays bounded across extended tasks.
  • The structured Chain-of-Thought trace, especially the failed-task feedback component, could be repurposed as a human-readable explanation of why a robot chose an action, but the paper does not explore interpretability.
  • A direct testable extension would be to systematically degrade instructions (missing object names, ambiguous pronouns) and vary graph update frequency, isolating whether robustness comes from the reasoning module, the graph, or their interaction.
  • Since the graph includes robot joint configurations, the approach might transfer across robot embodiments, but that transferability claim is an inference, not a result reported here.
Share X Bluesky LinkedIn Reddit HN

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 paper proposes GraphCoT-VLA, an end-to-end vision-language-action model that couples a structured Chain-of-Thought reasoning module (high-level task understanding, failed-task feedback, and low-level imaginative reasoning about future object positions/robot actions) with a real-time updatable 3D Pose-Object graph capturing robot joints and object topology, together with a dropout hybrid reasoning strategy for efficient control. The abstract claims significant improvements over existing methods in task success rate and response speed, plus strong generalization and robustness in open environments and under ambiguous instructions, based on multiple real-world robotic manipulation tasks.

Significance. If the claimed results hold, this would be a useful advance: integrating structured spatial reasoning via an updatable 3D graph into VLA control is an important direction, and the explicit handling of ambiguous instructions addresses a recognized limitation. The paper promises real-world validation and an efficiency-oriented reasoning strategy. However, because only the abstract was available, the empirical claims, architectural specifications, and ablations cannot be assessed. The significance is therefore conditional on the full manuscript providing quantitative support and detailed component analysis.

major comments (3)
  1. [Abstract, final paragraph] The central claim that GraphCoT-VLA 'significantly outperforms existing methods' is not verifiable from the abstract: no task list, baseline identities, success-rate values, effect sizes, error bars, or statistical tests are provided. This is load-bearing for the paper's main empirical assertion. The full manuscript must include a complete experimental protocol and quantitative comparisons.
  2. [Abstract, 3D Pose-Object graph] The real-time updatable 3D Pose-Object graph is a key novel component, but the abstract does not state whether it is constructed from raw sensor observations (e.g., RGB-D and joint encoders) or relies on privileged simulator ground truth. It also gives no update rule, latency bound, or error characterization. If the graph uses privileged pose information or accumulates pose/topology errors during manipulation, the claimed generalization to open environments would be unsupported. Please specify the sensor model, update mechanism, and error behavior.
  3. [Abstract, structured CoT module] The structured CoT includes 'failed task feedback' and 'imaginative reasoning about future object positions and robot actions.' If evaluation allows retries, or if failed-task feedback is provided during test rollouts, task solutions could leak into the reasoning process and inflate success rates. The abstract does not describe train/test separation or how future-position reasoning is supervised. This is a correctness risk that must be clarified for the central claim to be credible.
minor comments (3)
  1. [Abstract, grammar] In the sentence 'We further integrates a dropout hybrid reasoning strategy', the verb should be 'integrate' to agree with 'We'.
  2. [Abstract, terminology] 'Open environments' and 'uncertain instructions' need operational definitions in the full manuscript so that the claimed generalization and robustness can be evaluated.
  3. [Abstract, dropout hybrid reasoning] The 'dropout hybrid reasoning strategy' is mentioned as a contribution but its role in efficiency and accuracy is not described; a sentence explaining what is hybrid and what is dropped would be helpful.

Circularity Check

0 steps flagged

Abstract-only review: no derivation chain or fitted-input-to-prediction step is present, so no circularity is identifiable.

full rationale

The review input is the abstract only; the paper's equations, training procedures, evaluation protocol, and references are unavailable. A circularity finding requires quoting a specific reduction — e.g., a quantity defined in terms of the outcome it claims to predict, a fitted parameter renamed as a prediction, or a load-bearing argument that rests entirely on a self-citation. None of these can be exhibited from the abstract. The abstract's claim is an empirical one: GraphCoT-VLA outperforms existing methods in success rate and speed on real-world tasks. That is a benchmark-comparison claim, not a derivation that reduces to its inputs by construction. The phrase 'failed task feedback' could raise an experimental-leakage concern if evaluation allows retries on the same rollout, but the abstract describes no such protocol, and leakage is not the same as definitional circularity. No self-citations are even visible in the abstract. Therefore the honest finding is no significant circularity, score 0. Concerns about unvalidated 3D graph accuracy or ambiguity in the evaluation protocol are correctness/empirical-validity risks, not circularity, and are out of scope for this pass.

Axiom & Free-Parameter Ledger

1 free parameters · 3 axioms · 1 invented entities

Since only the abstract is available, the ledger records the assumptions that must hold for the abstract's claims to be true. No specific fitted parameter values, loss weights, or architecture hyperparameters are disclosed. The 3D Pose-Object graph is the only clearly introduced representational entity.

free parameters (1)
  • Learned model weights and training hyperparameters
    The central performance claim depends on training a VLA with unspecified architecture details such as CoT module, graph network, and dropout schedule. The abstract discloses no parameter counts, loss weights, or tuning procedure.
axioms (3)
  • domain assumption A real-time 3D Pose-Object graph can be estimated from raw observations with enough accuracy to guide manipulation.
    The abstract asserts this component but does not state whether it uses ground-truth poses, an external estimator, or learned prediction, nor its update rate and error behavior.
  • domain assumption Structured chain-of-thought reasoning with failed-task feedback improves planning rather than distracting from it.
    The abstract claims this improves interpretation of ambiguous instructions, but no ablation evidence is visible.
  • domain assumption The reported real-world tasks and baselines are representative and fairly compared.
    The superiority claim depends on fair baselines and open-environment test protocols; the abstract provides no task list, baseline names, or statistical details.
invented entities (1)
  • 3D Pose-Object graph no independent evidence
    purpose: Real-time graph encoding of robot joint positions and topological relations between objects to inform action selection.
    This is an internal representation introduced by the paper; no external, falsifiable handle such as a measurable physical quantity is provided in the abstract. Its validity is only evidenced indirectly by task success claims.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of GraphCoT-VLA: A 3D Spatial-Aware Reasoning Vision-Language-Action Model for Robotic Manipulation with Ambiguous Instructions." pith.science (2026). https://pith.science/paper/WC7ENKS4

@misc{pith2026250807650,
  author       = {Pith},
  title        = {Pith review of: GraphCoT-VLA: A 3D Spatial-Aware Reasoning Vision-Language-Action Model for Robotic Manipulation with Ambiguous Instructions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WC7ENKS4}},
  note         = {Machine review of arXiv:2508.07650}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Vision-language-action models have emerged as a crucial paradigm in robotic manipulation. However, existing VLA models exhibit notable limitations in handling ambiguous language instructions and unknown environmental states. Furthermore, their perception is largely constrained to static two-dimensional observations, lacking the capability to model three-dimensional interactions between the robot and its environment. To address these challenges, this paper proposes GraphCoT-VLA, an efficient end-to-end model. To enhance the model's ability to interpret ambiguous instructions and improve task planning, we design a structured Chain-of-Thought reasoning module that integrates high-level task understanding and planning, failed task feedback, and low-level imaginative reasoning about future object positions and robot actions. Additionally, we construct a real-time updatable 3D Pose-Object graph, which captures the spatial configuration of robot joints and the topological relationships between objects in 3D space, enabling the model to better understand and manipulate their interactions. We further integrates a dropout hybrid reasoning strategy to achieve efficient control outputs. Experimental results across multiple real-world robotic tasks demonstrate that GraphCoT-VLA significantly outperforms existing methods in terms of task success rate and response speed, exhibiting strong generalization and robustness in open environments and under uncertain instructions.

discussion (0)

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

Forward citations

Cited by 9 Pith papers

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

  1. DSSP: Diffusion State Space Policy with Full-History Encoding

    cs.RO 2026-05 conditional novelty 7.0

    DSSP is a history-conditioned diffusion state space policy that uses SSMs to encode full observation streams with an auxiliary dynamics objective and hierarchical fusion, achieving SOTA results with reduced model size...

  2. RoboEvolve: Co-Evolving Planner-Simulator for Robotic Manipulation with Limited Data

    cs.RO 2026-05 unverdicted novelty 6.0

    A co-evolutionary VLM-VGM loop on 500 unlabeled images raises planner success by 30 points and simulator success by 48 percent while beating fully supervised baselines.

  3. TriRelVLA: Triadic Relational Structure for Generalizable Embodied Manipulation

    cs.CV 2026-05 unverdicted novelty 6.0

    TriRelVLA introduces triadic object-hand-task relational representations and a task-grounded graph transformer with a relational bottleneck to improve generalization in robotic manipulation across scenes, objects, and tasks.

  4. ProGAL-VLA: Grounded Alignment through Prospective Reasoning in Vision-Language-Action Models

    cs.RO 2026-04 unverdicted novelty 6.0

    ProGAL-VLA uses 3D graphs, symbolic sub-goals, and a Grounding Alignment Contrastive loss to ground actions on verified embeddings, raising robustness from 30.3% to 71.5% and ambiguity AUROC to 0.81 on robotic benchmarks.

  5. VLSA: Vision-Language-Action Models with Plug-and-Play Safety Constraint Layer

    cs.RO 2025-12 conditional novelty 6.0

    AEGIS wraps VLA robot policies in a CBF-based safety layer that uses VLM obstacle identification, raising collision avoidance from 18.69% to 77.85% and task success by 17.25 points on the new SafeLIBERO benchmark.

  6. VistaVLA: Geometry- and Semantic-Aware 3D Gaussian-Grounded VLA for Robotic Manipulation

    cs.RO 2026-07 conditional novelty 5.5

    VistaVLA lifts multi-view vision-language features into 3D Gaussians, compresses them 99% via Merge-then-Query, and improves real-robot manipulation success by ~23% over baselines.

  7. VistaVLA: Geometry- and Semantic-Aware 3D Gaussian-Grounded VLA for Robotic Manipulation

    cs.RO 2026-07 conditional novelty 5.0

    VistaVLA lifts 2D semantic features into 3D Gaussians and compresses them to 64 tokens, improving robot manipulation success by 22.8% over a 2D VLA baseline in real-world tasks.

  8. Coarse-to-Control: Action-Token Planning for Vision-Language-Action Models

    cs.RO 2026-06 unverdicted novelty 5.0

    Coarse-to-Control adds planning via coarse action tokens in the same vocabulary as control actions, improving VLA performance on long-horizon manipulation tasks.

  9. X-Imitator: Spatial-Aware Imitation Learning via Bidirectional Action-Pose Interaction

    cs.RO 2026-05 unverdicted novelty 5.0

    X-Imitator is a bidirectional action-pose interaction framework for spatial-aware imitation learning that outperforms vanilla policies and explicit pose guidance on 24 simulated and 3 real-world robotic tasks.

Reference graph

Works this paper leans on

32 extracted references · 4 canonical work pages · cited by 8 Pith papers

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Bai, S.; Chen, K.; Liu, X.; Wang, J.; Ge, W.; Song, S.; Dang, K.; Wang, P.; Wang, S.; Tang, J.; et al. 2025. Qwen2. 5-vl technical report. arXiv preprint arXiv:2502.13923

    Pith/arXiv arXiv 2025

  4. [4]

    S.; Kolesnikov, A.; Wang, X.; Salz, D.; Neumann, M.; Alabdulmohsin, I.; Tschannen, M.; Bugliarello, E.; et al

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

    Pith/arXiv arXiv 2024

  5. [5]

    Black, K.; Brown, N.; Driess, D.; Esmail, A.; Equi, M.; Finn, C.; Fusai, N.; Groom, L.; Hausman, K.; Ichter, B.; et al. 2024. pi\_0 : A Vision-Language-Action Flow Model for General Robot Control. arXiv preprint arXiv:2410.24164

    Pith/arXiv arXiv 2024

  6. [6]

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

    Pith/arXiv arXiv 2022

  7. [7]

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

    Pith/arXiv arXiv 2024

  8. [8]

    Chen, W.; Belkhale, S.; Mirchandani, S.; Mees, O.; Driess, D.; Pertsch, K.; and Levine, S. 2025. Training Strategies for Efficient Embodied Reasoning. arXiv preprint arXiv:2505.08243

    Pith/arXiv arXiv 2025

  9. [9]

    Cheng, T.; Song, L.; Ge, Y.; Liu, W.; Wang, X.; and Shan, Y. 2024. Yolo-world: Real-time open-vocabulary object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 16901--16911

    2024

  10. [10]

    Chi, C.; Xu, Z.; Feng, S.; Cousineau, E.; Du, Y.; Burchfiel, B.; Tedrake, R.; and Song, S. 2023. Diffusion policy: Visuomotor policy learning via action diffusion. The International Journal of Robotics Research, 02783649241273668

    2023

  11. [11]

    U.; Akram, W.; Saoud, L

    Din, M. U.; Akram, W.; Saoud, L. S.; Rosell, J.; and Hussain, I. 2025. Vision Language Action Models in Robotic Manipulation: A Systematic Review. arXiv preprint arXiv:2507.10672

    arXiv 2025

  12. [12]

    Dosovitskiy, A.; Beyer, L.; Kolesnikov, A.; Weissenborn, D.; Zhai, X.; Unterthiner, T.; Dehghani, M.; Minderer, M.; Heigold, G.; Gelly, S.; et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929

    Pith/arXiv arXiv 2020

  13. [13]

    J.; Pertsch, K.; Karamcheti, S.; Xiao, T.; Balakrishna, A.; Nair, S.; Rafailov, R.; Foster, E.; Lam, G.; Sanketi, P.; et al

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

    Pith/arXiv arXiv 2024

  14. [14]

    T.; Ben-Hamu, H.; Nickel, M.; and Le, M

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

    Pith/arXiv arXiv 2022

  15. [15]

    Liu, S.; Wu, L.; Li, B.; Tan, H.; Chen, H.; Wang, Z.; Xu, K.; Su, H.; and Zhu, J. 2024. Rdt-1b: a diffusion foundation model for bimanual manipulation. arXiv preprint arXiv:2410.07864

    Pith/arXiv arXiv 2024

  16. [16]

    Mao, A.; Mohri, M.; and Zhong, Y. 2023. Cross-entropy loss functions: Theoretical analysis and applications. In International conference on Machine learning, 23803--23828. pmlr

    work page 2023

  17. [17]

    O’Neill, A.; Rehman, A.; Maddukuri, A.; Gupta, A.; Padalkar, A.; Lee, A.; Pooley, A.; Gupta, A.; Mandlekar, A.; Jain, A.; et al. 2024. Open x-embodiment: Robotic learning datasets and rt-x models: Open x-embodiment collaboration 0. In 2024 IEEE International Conference on Robotics and Automation (ICRA), 6892--6903. IEEE

    2024

  18. [18]

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

    Pith/arXiv arXiv 2025

  19. [19]

    C.; Hagenbuchner, M.; and Monfardini, G

    Scarselli, F.; Gori, M.; Tsoi, A. C.; Hagenbuchner, M.; and Monfardini, G. 2008. The graph neural network model. IEEE transactions on neural networks, 20(1): 61--80

    2008

  20. [20]

    M.; Ghosh, D.; Walke, H.; Pertsch, K.; Black, K.; Mees, O.; Dasari, S.; Hejna, J.; Kreiman, T.; Xu, C.; et al

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

    Pith/arXiv arXiv 2024

  21. [21]

    N.; Kaiser, .; and Polosukhin, I

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

    2017

  22. [22]

    Wang, S. 2025. Roboflamingo-plus: Fusion of depth and rgb perception with vision-language models for enhanced robotic manipulation. arXiv preprint arXiv:2503.19510

    Pith/arXiv arXiv 2025

  23. [23]

    V.; Zhou, D.; et al

    Wei, J.; Wang, X.; Schuurmans, D.; Bosma, M.; Xia, F.; Chi, E.; Le, Q. V.; Zhou, D.; et al. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35: 24824--24837

    2022

  24. [24]

    Wen, J.; Zhu, M.; Zhu, Y.; Tang, Z.; Li, J.; Zhou, Z.; Li, C.; Liu, X.; Peng, Y.; Shen, C.; et al. 2024. Diffusion-VLA: Generalizable and Interpretable Robot Foundation Model via Self-Generated Reasoning. arXiv preprint arXiv:2412.03293

    Pith/arXiv arXiv 2024

  25. [25]

    Wen, J.; Zhu, Y.; Li, J.; Tang, Z.; Shen, C.; and Feng, F. 2025 a . Dexvla: Vision-language model with plug-in diffusion expert for general robot control. arXiv preprint arXiv:2502.05855

    Pith/arXiv arXiv 2025

  26. [26]

    Wen, J.; Zhu, Y.; Li, J.; Zhu, M.; Tang, Z.; Wu, K.; Xu, Z.; Liu, N.; Cheng, R.; Shen, C.; et al. 2025 b . Tinyvla: Towards fast, data-efficient vision-language-action models for robotic manipulation. IEEE Robotics and Automation Letters

    work page 2025

  27. [27]

    Wu, H.; Jing, Y.; Cheang, C.; Chen, G.; Xu, J.; Li, X.; Liu, M.; Li, H.; and Kong, T. 2023. Unleashing large-scale video generative pre-training for visual robot manipulation. arXiv preprint arXiv:2312.13139

    Pith/arXiv arXiv 2023

  28. [28]

    Zawalski, M.; Chen, W.; Pertsch, K.; Mees, O.; Finn, C.; and Levine, S. 2024. Robotic control via embodied chain-of-thought reasoning. arXiv preprint arXiv:2407.08693

    Pith/arXiv arXiv 2024

  29. [29]

    Zhang, Y.; and Yan, J. 2023. Crossformer: Transformer utilizing cross-dimension dependency for multivariate time series forecasting. In The eleventh international conference on learning representations

    work page 2023

  30. [30]

    J.; Fu, Z.; Zhang, Z.; Wu, Y.; Li, Z.; Ma, Q.; Han, S.; Finn, C.; et al

    Zhao, Q.; Lu, Y.; Kim, M. J.; Fu, Z.; Zhang, Z.; Wu, Y.; Li, Z.; Ma, Q.; Han, S.; Finn, C.; et al. 2025. Cot-vla: Visual chain-of-thought reasoning for vision-language-action models. In Proceedings of the Computer Vision and Pattern Recognition Conference, 1702--1713

    work page 2025

  31. [31]

    Z.; Kumar, V.; Levine, S.; and Finn, C

    Zhao, T. Z.; Kumar, V.; Levine, S.; and Finn, C. 2023. Learning fine-grained bimanual manipulation with low-cost hardware. arXiv preprint arXiv:2304.13705

    Pith/arXiv arXiv 2023

  32. [32]

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

    2023

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.