pith. sign in

arxiv: 2607.02503 · v1 · pith:Y4ZLLLQPnew · submitted 2026-07-02 · 💻 cs.RO

VT-WAM: Visual-Tactile World Action Model for Contact-Rich Manipulation

Shuai Tian , Yupeng Zheng , Yuhang Zheng , Songen Gu , Yujie Zang , Yuxing Qin , Weize Li , Haoran Li
show 2 more authors
Wenchao Ding Dongbin Zhao
This is my paper

Pith reviewed 2026-07-03 10:30 UTC · model grok-4.3

classification 💻 cs.RO
keywords visual-tactile policycontact-rich manipulationflow matchingtactile deformation predictionattention guidanceworld action modelrobot learningaction prediction
0
0 comments X

The pith

VT-WAM jointly predicts visual futures, tactile deformations, and actions in a flow-matching model to handle contact-rich robot tasks.

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

The paper introduces VT-WAM to handle contact-rich manipulation, where policies must respond to local deformation, pressure, slip, and friction that visual observations often miss because the signals are temporally sparse. It trains a single model to predict future visual scenes, tactile sensor deformations over time, and robot actions together inside a flow-matching framework. Two new components are an asymmetric mixture-of-transformers attention that links a starting visual frame to ongoing tactile changes, and contact-gated attention guidance that directs action decisions toward tactile evidence specifically during contact phases. On six real-world tasks the model reaches 71.67 percent average success, exceeding two prior visual-tactile baselines, with ablations showing that deformation modeling and contact guidance each contribute. A sympathetic reader would care because many practical manipulation problems fail when policies cannot anticipate how objects and sensors will move and deform on contact.

Core claim

VT-WAM is a Visual-Tactile World Action Model that jointly learns future visual prediction, tactile deformation prediction, and action prediction within a unified flow matching framework. It introduces Asymmetric Mixture-of-Transformers attention to bridge a first-frame visual anchor with temporal tactile dynamics, and contact-gated Action-Visual-Tactile Attention Guidance to encourage action queries to rely on tactile evidence during contact phases. Across six real-world contact-rich manipulation tasks, VT-WAM achieves a 71.67% average success rate, outperforming Fast-WAM by 26.67% and OmniVTLA by 35.84%. Ablations demonstrate that modeling tactile deformation dynamics and guiding contact-p

What carries the argument

Asymmetric Mixture-of-Transformers (MoT) attention and contact-gated Action-Visual-Tactile Attention Guidance (AVTAG) inside a unified flow-matching generative model for joint visual, tactile deformation, and action prediction.

Load-bearing premise

The Asymmetric Mixture-of-Transformers attention and contact-gated AVTAG will produce action predictions that benefit from modeling tactile deformation dynamics when trained inside the flow-matching framework on the evaluated tasks and hardware.

What would settle it

Training and testing a version of VT-WAM on the same six tasks with the tactile deformation prediction head removed and checking whether success rates fall to the level of the prior baselines would test whether the deformation modeling component is necessary.

Figures

Figures reproduced from arXiv: 2607.02503 by Dongbin Zhao, Haoran Li, Shuai Tian, Songen Gu, Weize Li, Wenchao Ding, Yuhang Zheng, Yujie Zang, Yupeng Zheng, Yuxing Qin.

Figure 1
Figure 1. Figure 1: Sparse tactile dynamics provide decisive evidence for contact￾rich manipulation. (a) Across six real-world tasks, tactile responses appear mainly around short contact events, making the informative signal tem￾porally sparse. (b) By coupling action prediction with tactile deformation dynamics, VT-WAM improves the average success rate from 45.00% with Fast-WAM to 71.67%, with consistent gains on both surface… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of VT-WAM. (a) Joint visual-tactile-action flow matching with three modality-specific experts connected by Asymmetric MoT Attention. (b) Attention masks in Asymmetric MoT Attention during training and inference. (c) Contact-gated AVTAG applies a training-only hinge ranking loss that encourages action queries to prioritize tactile evidence during contact phases. this concatenated token sequence: P … view at source ↗
Figure 3
Figure 3. Figure 3: Real-world experimental platform. The setup uses a 7-DoF xArm7 robot with a Robotiq 2F-85 gripper, a wrist camera, and paired gripper-mounted Xense tactile sensors. The scene includes the representative objects used in our experiments. These two quantities are then normalized into relative visual and tactile attention weights: pv(r) = αv(r) αv(r) + αt(r) , pt(r) = αt(r) αv(r) + αt(r) . AVTAG applies this g… view at source ↗
Figure 4
Figure 4. Figure 4: Overview of real-world contact-rich manipulation tasks. We evaluate VT-WAM on six real-world tasks covering two interaction regimes: surface-interaction tasks and constrained insertion tasks. TABLE I SUCCESS RATES ON REAL-WORLD CONTACT-RICH TASKS. Method Surface-Interaction Tasks Constrained Insertion Tasks Average Wipe Board Wipe Vase Peel Cucumber Avg. Insert Plug Swipe Card Insert Tube Avg. DP + Tactile… view at source ↗
Figure 5
Figure 5. Figure 5: Visual-Tactile Prediction Results across Six Tasks. For visualization, VT-WAM predicts wrist camera observations together with tactile deformation fields. Blue denotes ground truth, and orange indicates prediction. • RDP [3]: a reactive visual-tactile policy that uses tactile feedback for online action refinement. • π0.5 [2]: a general vision-language-action policy with￾out tactile input. • OmniVTLA [5]: a… view at source ↗
Figure 6
Figure 6. Figure 6: AVTAG promotes tactile attention for contact recovery during vase wiping. The red and blue curves denote relative tactile and visual attention weights pt and pv from the action expert, and the dashed curve denotes the contact force |Fz| for visualization only. The wrist camera view is the only visual input available to the policy, while the side view is shown only for visualization. When the supporting pla… view at source ↗
read the original abstract

Contact-rich manipulation requires policies to react to local deformation, pressure, slip, and friction, yet these cues are temporally sparse and often invisible in visual observations. Existing visual-tactile policies usually feed tactile observations directly into action prediction, but rarely model tactile deformation dynamics during action generation. In this paper, we introduce VT-WAM, a Visual-Tactile World Action Model that jointly learns future visual prediction, tactile deformation prediction, and action prediction within a unified flow matching framework. In particular, VT-WAM introduces (1) Asymmetric Mixture-of-Transformers (MoT) attention to bridge a first-frame visual anchor with temporal tactile dynamics, and (2) contact-gated Action-Visual-Tactile Attention Guidance (AVTAG) to encourage action queries to rely on tactile evidence during contact phases. Across six real-world contact-rich manipulation tasks, VT-WAM achieves a 71.67% average success rate, outperforming Fast-WAM by 26.67% and OmniVTLA by 35.84%. Ablations demonstrate that modeling tactile deformation dynamics and guiding contact-phase tactile attention are both important for contact-rich tasks. Project website: https://vt-wam.github.io/.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 3 minor

Summary. The manuscript introduces VT-WAM, a Visual-Tactile World Action Model that jointly predicts future visual observations, tactile deformations, and actions inside a single flow-matching framework. It proposes two architectural components—Asymmetric Mixture-of-Transformers attention that anchors on the first visual frame while attending to temporal tactile dynamics, and contact-gated AVTAG that routes action queries to tactile evidence only during contact phases—and evaluates the model on six real-world contact-rich manipulation tasks, reporting a 71.67 % mean success rate that exceeds Fast-WAM by 26.67 % and OmniVTLA by 35.84 %. Ablations are cited to show that both tactile-deformation modeling and contact-phase guidance contribute measurably to performance.

Significance. If the reported gains and ablation results hold under full experimental scrutiny, the work supplies concrete evidence that explicit, dynamics-aware tactile modeling inside a generative action framework improves reliability on contact-rich tasks where visual cues alone are insufficient. The joint flow-matching objective and the two attention mechanisms constitute a reusable design pattern that could be adopted by other multi-modal manipulation pipelines.

minor comments (3)
  1. [Abstract] The abstract states that ablations confirm the importance of the two proposed components, yet no quantitative drops (e.g., success-rate deltas or per-task breakdowns) are supplied; these numbers should appear in the main results table or a dedicated ablation subsection.
  2. The description of the Asymmetric MoT attention and contact-gated AVTAG would benefit from an explicit diagram or pseudocode block showing how the first-frame visual anchor is injected and how the contact gate is computed from tactile signals.
  3. Task definitions, success criteria, and hardware specifications (sensor models, robot platform, force thresholds) are referenced only at a high level; a concise table listing these details would improve reproducibility.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary of VT-WAM and the recommendation for minor revision. No specific major comments were provided in the report, so we have no individual points requiring rebuttal or revision at this stage. We remain available to address any additional feedback or minor clarifications during the revision process.

Circularity Check

0 steps flagged

No significant circularity; empirical claims rest on external data and ablations

full rationale

The paper reports an empirical performance result (71.67% success rate on six real-world tasks) from training a flow-matching model with Asymmetric MoT attention and contact-gated AVTAG. No equations, parameter-fitting procedures, or derivation steps are described that would reduce the reported success rates or ablation outcomes to quantities defined by the same fitted parameters. The central claims are supported by direct experimental evidence (ablations) rather than by self-referential definitions or imported uniqueness theorems. This is the normal case of a self-contained empirical robotics paper.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities beyond the high-level model description can be extracted. Typical deep-learning models contain many fitted weights, but none are itemized here.

pith-pipeline@v0.9.1-grok · 5775 in / 1090 out tokens · 49117 ms · 2026-07-03T10:30:21.722776+00:00 · methodology

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

34 extracted references · 18 canonical work pages · 7 internal anchors

  1. [1]

    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

  2. [2]

    $\pi_{0.5}$: a Vision-Language-Action Model with Open-World Generalization

    K. Black, N. Brown, J. Darpinian, K. Dhabalia, D. Driess, A. Es- mail, M. Equi, C. Finn, N. Fusaiet al., “π 0.5: A Vision-Language- Action Model with Open-World Generalization,”arXiv preprint arXiv:2504.16054, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  3. [3]

    Reactive Diffusion Policy: Slow-Fast Visual-Tactile Policy Learning for Contact-Rich Manipulation,

    H. Xue, J. Ren, W. Chen, G. Zhang, Y . Fang, G. Gu, H. Xu, and C. Lu, “Reactive Diffusion Policy: Slow-Fast Visual-Tactile Policy Learning for Contact-Rich Manipulation,” inProceedings of Robotics: Science and Systems (RSS), 2025

    2025

  4. [4]

    VLA-Touch: Enhancing Vision-Language-Action Model with Dual-Level Tactile Feedback,

    J. Bi, K. Y . Ma, C. Hao, M. S. Zheng, and H. Soh, “VLA-Touch: Enhancing Vision-Language-Action Model with Dual-Level Tactile Feedback,”IEEE Robotics and Automation Letters, 2026

    2026

  5. [5]

    Cheng, Y

    Z. Cheng, Y . Zhang, W. Zhang, H. Li, K. Wang, L. Song, and H. Zhang, “OmniVTLA: Vision-Tactile-Language-Action Model with Semantic-Aligned Tactile Sensing,”arXiv preprint arXiv:2508.08706, 2025

    work page arXiv 2025

  6. [6]

    VisuoTactile-RL: Learning multimodal manipulation policies with deep reinforcement learning,

    J. Hansen, F. Hogan, D. Rivkin, D. Meger, M. Jenkin, and G. Dudek, “VisuoTactile-RL: Learning multimodal manipulation policies with deep reinforcement learning,” in2022 International Conference on Robotics and Automation (ICRA). IEEE, 2022, pp. 8298–8304

    2022

  7. [7]

    Unified World Models: Coupling Video and Action Diffusion for Pretraining on Large Robotic Datasets,

    C. Zhu, R. Yu, S. Feng, B. Burchfiel, P. Shah, and A. Gupta, “Unified World Models: Coupling Video and Action Diffusion for Pretraining on Large Robotic Datasets,” inProceedings of Robotics: Science and Systems (RSS), 2025

    2025

  8. [8]

    Causal World Modeling for Robot Control,

    L. Li, Q. Zhang, Y . Luo, S. Yang, R. Wang, F. Han, M. Yu, Z. Gao, N. Xue, X. Zhuet al., “Causal World Modeling for Robot Control,” inProceedings of Robotics: Science and Systems (RSS), 2026

    2026

  9. [9]

    Fast-WAM: Do World Action Models Need Test-time Future Imagination?

    T. Yuan, Z. Dong, Y . Liu, and H. Zhao, “Fast-W AM: Do World Action Models Need Test-Time Future Imagination?”arXiv preprint arXiv:2603.16666, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  10. [10]

    Tactile-conditioned diffusion policy for force-aware robotic manipulation.arXiv preprint arXiv:2510.13324, 2025

    E. Helmut, N. Funk, T. Schneider, C. de Farias, and J. Peters, “Tactile- Conditioned Diffusion Policy for Force-Aware Robotic Manipulation,” arXiv preprint arXiv:2510.13324, 2025

    work page arXiv 2025

  11. [11]

    TacDiffusion: Force-Domain Diffusion Policy for Precise Tactile Manipulation,

    Y . Wu, Z. Chen, F. Wu, L. Chen, L. Zhanget al., “TacDiffusion: Force-Domain Diffusion Policy for Precise Tactile Manipulation,” in Proceedings of the IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 11 831–11 837

    2025

  12. [12]

    PolyTouch: A Robust Multi-Modal Tactile Sensor for Contact-Rich Manipulation Using Tactile-Diffusion Policies,

    J. Zhao, N. Kuppuswamy, S. Feng, B. Burchfiel, and E. Adelson, “PolyTouch: A Robust Multi-Modal Tactile Sensor for Contact-Rich Manipulation Using Tactile-Diffusion Policies,” inProceedings of the IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 104–110

    2025

  13. [13]

    Zhang, C

    D. Zhang, C. Yuan, C. Wen, H. Zhang, J. Zhao, and Y . Gao, “KineDex: Learning Tactile-Informed Visuomotor Policies via Kinesthetic Teach- ing for Dexterous Manipulation,”arXiv preprint arXiv:2505.01974, 2025

    work page arXiv 2025

  14. [14]

    Force Policy: Learning hybrid force-position control policy under interaction frame for contact-rich manipulation,

    H. Fang, S. Tang, M. Mei, H. Qin, Z. He, J. Chen, Y . Feng, C. Wang, W. Liu, Z. Heet al., “Force Policy: Learning hybrid force-position control policy under interaction frame for contact-rich manipulation,” inProceedings of Robotics: Science and Systems (RSS), 2026

    2026

  15. [15]

    Master Micro Residual Correction with Adaptive Tactile Fusion and Force-Mixed Control for Contact-Rich Manipulation,

    X. Li, Y . Xie, H. Liu, W. Hou, G. Chen, S. Li, and W. Ding, “Master Micro Residual Correction with Adaptive Tactile Fusion and Force-Mixed Control for Contact-Rich Manipulation,”arXiv preprint arXiv:2603.15152, 2026

    work page arXiv 2026

  16. [16]

    BiTLA: A Bimanual Tactile-Language-Action Model for Contact-Rich Robotic Manipulation,

    S. Yang, H. Li, J. Hu, S. Zhang, G. Yao, Z. Ni, and B. Fang, “BiTLA: A Bimanual Tactile-Language-Action Model for Contact-Rich Robotic Manipulation,” inProceedings of the 1st International Workshop on Multi-Sensorial Media and Applications, 2025, pp. 12–17

    2025

  17. [17]

    Zhang, P

    C. Zhang, P. Hao, X. Cao, X. Hao, S. Cui, and S. Wang, “VTLA: Vision-Tactile-Language-Action Model with Preference Learning for Insertion Manipulation,”arXiv preprint arXiv:2505.09577, 2025

    work page arXiv 2025

  18. [18]

    Tactile-VLA: Unlocking vision-language-action model’s physical knowledge for tactile generalization.arXiv preprint arXiv:2507.09160,

    J. Huang, S. Wang, F. Lin, Y . Hu, C. Wen, and Y . Gao, “Tactile-VLA: Unlocking Vision-Language-Action Model’s Physical Knowledge for Tactile Generalization,”arXiv preprint arXiv:2507.09160, 2025

    work page arXiv 2025

  19. [19]

    Tactile-Force Alignment in Vision-Language-Action Models for Force-Aware Manipulation,

    Y . Huang, P. Lin, W. Li, D. Li, J. Li, J. Jiang, C. Xiao, and Z. Jiao, “Tactile-Force Alignment in Vision-Language-Action Models for Force-Aware Manipulation,”arXiv preprint arXiv:2601.20321, 2026

    work page arXiv 2026

  20. [20]

    Visuo-Tactile World Models,

    C. Higuera, S. Arnaud, B. Boots, M. Mukadam, F. R. Hogan, and F. Meier, “Visuo-Tactile World Models,”arXiv preprint arXiv:2602.06001, 2026

    work page arXiv 2026

  21. [21]

    Zheng, S

    Y . Zheng, S. Gu, W. Li, Y . Zheng, Y . Zang, S. Tian, X. Li, C. Hao, C. Gao, S. Liuet al., “OmniVTA: Visuo-Tactile World Modeling for Contact-Rich Robotic Manipulation,”arXiv preprint arXiv:2603.19201, 2026

    work page arXiv 2026

  22. [22]

    World Action Models: The Next Frontier in Embodied AI

    S. Wang, J. Shi, Z. Fu, X. He, F. Liu, C. Yang, Y . Zhou, Z. Fei, J. Gong, J. Fuet al., “World Action Models: The Next Frontier in Embodied AI,”arXiv preprint arXiv:2605.12090, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  23. [23]

    Learning Universal Policies via Text-Guided Video Generation,

    Y . Du, S. Yang, B. Dai, H. Dai, O. Nachum, J. Tenenbaum, D. Schu- urmans, and P. Abbeel, “Learning Universal Policies via Text-Guided Video Generation,”Advances in Neural Information Processing Sys- tems, vol. 36, pp. 9156–9172, 2023

    2023

  24. [24]

    Video Language Planning,

    Y . Du, S. Yang, P. Florence, F. Xia, A. Wahid, P. Sermanet, T. Yu, P. Abbeel, J. B. Tenenbaum, L. Kaelblinget al., “Video Language Planning,” inProceedings of the International Conference on Learning Representations (ICLR), vol. 2024, 2024, pp. 31 138–31 155

    2024

  25. [25]

    RoboEnvision: A Long-Horizon Video Generation Model for Multi-Task Robot Ma- nipulation,

    L. Yang, Y . Bai, G. Eskandar, F. Shen, M. Altillawi, D. Chen, S. Majumder, Z. Liu, G. Kutyniok, and A. Valada, “RoboEnvision: A Long-Horizon Video Generation Model for Multi-Task Robot Ma- nipulation,” inProceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2025, pp. 21 281–21 288

    2025

  26. [26]

    Say, Dream, and Act: Learning Video World Models for Instruction-Driven Robot Manipulation,

    S. Gu, Y . Cai, T. Wang, S. Wu, and Y . Fu, “Say, Dream, and Act: Learning Video World Models for Instruction-Driven Robot Manipulation,”arXiv preprint arXiv:2602.10717, 2026

    work page arXiv 2026

  27. [27]

    World Action Models are Zero-shot Policies

    S. Ye, Y . Ge, K. Zheng, S. Gao, S. Yu, G. Kurian, S. Indupuru, Y . L. Tan, C. Zhu, J. Xianget al., “World Action Models Are Zero-Shot Policies,”arXiv preprint arXiv:2602.15922, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  28. [28]

    Motus: A Unified Latent Action World Model

    H. Bi, H. Tan, S. Xie, Z. Wang, S. Huang, H. Liu, R. Zhao, Y . Feng, C. Xiang, Y . Ronget al., “Motus: A Unified Latent Action World Model,”arXiv preprint arXiv:2512.13030, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  29. [29]

    Gigaworld-policy: An efficient action-centered world–action model.arXiv preprint arXiv:2603.17240, 2026

    A. Ye, B. Wang, C. Ni, G. Huang, G. Zhao, H. Li, H. Li, J. Li, J. Lv, J. Liuet al., “GigaWorld-Policy: An Efficient Action-Centered World–Action Model,”arXiv preprint arXiv:2603.17240, 2026

    work page arXiv 2026

  30. [30]

    Wan: Open and Advanced Large-Scale Video Generative Models

    A. Wang, B. Ai, B. Wen, C. Mao, C.-W. Xie, D. Chen, F. Yu, H. Zhao, J. Yanget al., “Wan: Open and Advanced Large-Scale Video Generative Models,”arXiv preprint arXiv:2503.20314, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  31. [31]

    Attention Is All You Need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention Is All You Need,”Ad- vances in Neural Information Processing Systems (NeurIPS), vol. 30, 2017

    2017

  32. [32]

    Mixture-of- Transformers: A Sparse and Scalable Architecture for Multi-Modal Foundation Models,

    W. Liang, L. Yu, L. Luo, S. Iyer, N. Dong, C. Zhou, G. Ghosh, M. Lewis, W.-t. Yih, L. Zettlemoyer, and X. V . Lin, “Mixture-of- Transformers: A Sparse and Scalable Architecture for Multi-Modal Foundation Models,”Transactions on Machine Learning Research, 2025

    2025

  33. [33]

    exUMI: Extensible Robot Teaching System with Action-Aware Task-Agnostic Tactile Representation,

    Y . Xu, L. Wei, P. An, Q. Zhang, and Y .-L. Li, “exUMI: Extensible Robot Teaching System with Action-Aware Task-Agnostic Tactile Representation,” inProceedings of the Conference on Robot Learning (CoRL), 2025

    2025

  34. [34]

    Unified Video Action Model

    S. Li, Y . Gao, D. Sadigh, and S. Song, “Unified Video Action Model,” arXiv preprint arXiv:2503.00200, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025