pith. sign in

arxiv: 2601.20239 · v6 · submitted 2026-01-28 · 💻 cs.RO

TouchGuide: Inference-Time Steering of Visuomotor Policies via Touch Guidance

Zhemeng Zhang , Jiahua Ma , Xincheng Yang , Xin Wen , Yuzhi Zhang , Boyan Li , Yiran Qin , Jin Liu
show 8 more authors
Can Zhao Li Kang Haoqin Hong Zhenfei Yin Philip Torr Hao Su Ruimao Zhang Daolin Ma
This is my paper

Pith reviewed 2026-05-16 11:03 UTC · model grok-4.3

classification 💻 cs.RO
keywords touch guidancevisuomotor policycontact-rich manipulationinference-time steeringtactile feedbackdiffusion policyphysical contact modelrobot manipulation
0
0 comments X

The pith

TouchGuide steers pre-trained visuomotor policies at inference time using a contact physical model to produce physically valid actions for contact-rich tasks.

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

The paper introduces TouchGuide as a two-stage method that first lets a pre-trained diffusion or flow-matching visuomotor policy generate coarse actions from visual input alone, then applies a task-specific Contact Physical Model to score and steer the remaining sampling steps toward actions that meet realistic physical contact constraints. This inference-time fusion occurs in a low-dimensional action space and relies on contrastive training of the model on limited expert demonstrations. The approach is paired with a data collection system that gathers reliable tactile signals affordably. A sympathetic reader would care because it improves success rates on challenging tasks such as shoe lacing and chip handover without requiring full retraining of the base policy.

Core claim

TouchGuide operates in two stages to guide a pre-trained diffusion or flow-matching visuomotor policy at inference time. First, the policy produces a coarse, visually-plausible action using only visual inputs during early sampling. Second, a task-specific Contact Physical Model provides tactile guidance to steer and refine the action, ensuring it aligns with realistic physical contact conditions. Trained through contrastive learning on limited expert demonstrations, the CPM provides a tactile-informed feasibility score to steer the sampling process toward refined actions that satisfy physical contact constraints.

What carries the argument

The Contact Physical Model (CPM), a task-specific module trained via contrastive learning on expert demonstrations that supplies tactile-informed feasibility scores to refine policy sampling toward physically valid contacts.

If this is right

  • TouchGuide consistently outperforms state-of-the-art visuo-tactile policies on five contact-rich manipulation tasks.
  • The method works with both diffusion-based and flow-matching visuomotor policies without modifying their training.
  • Steering happens only at inference time, preserving the original policy while adding tactile constraints.
  • TacUMI enables collection of high-quality tactile data at lower cost using rigid fingertips for direct feedback.
  • The low-dimensional action-space fusion reduces the need for end-to-end multimodal retraining.

Where Pith is reading between the lines

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

  • This separation of visual generation and tactile refinement could let teams reuse large-scale visual pre-training across many contact tasks by swapping only the lightweight CPM.
  • If the feasibility scoring proves robust, similar inference-time modules might be added for other modalities such as force or audio to further constrain sampling.
  • The approach opens a path to rapid adaptation on new hardware or objects by collecting a small expert set and training one new CPM rather than retraining the full policy.
  • In deployment, the method could lower the rate of physically unsafe actions by rejecting low-feasibility samples before execution.

Load-bearing premise

The task-specific Contact Physical Model trained on limited expert demonstrations will generalize to give accurate feasibility scores that correctly identify and steer toward physically valid contact actions in new situations.

What would settle it

Measuring that guided actions on a held-out contact-rich task violate physical constraints at a similar rate to the unguided baseline or produce no improvement in task success rate would falsify the central claim.

Figures

Figures reproduced from arXiv: 2601.20239 by Boyan Li, Can Zhao, Daolin Ma, Haoqin Hong, Hao Su, Jiahua Ma, Jin Liu, Li Kang, Philip Torr, Ruimao Zhang, Xincheng Yang, Xin Wen, Yiran Qin, Yuzhi Zhang, Zhemeng Zhang, Zhenfei Yin.

Figure 1
Figure 1. Figure 1: TacUMI is a low-cost yet high-precision handheld data collection system that provides direct tactile feedback through a rigid mechanical coupling. TouchGuide is a multi-modal fusion paradigm that steers a visuomotor policy via touch guidance during denoising or flow matching, producing actions that better adhere to contact physics without retraining the base policy. Abstract—Fine-grained and contact-rich m… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of TacUMI data collection system. (a) TacUMI (Collection-side) uses a Vive tracker for localization to obtain accurate end-effector poses, while the operator receives direct tactile feedback. (b) During policy inference, we use an execution-side device that is structurally identical to the collection-side TacUMI, coupled to different robot arms via an adapter. robotic manipulation, which is an eff… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of TouchGuide framework. (a) The architecture of the task-specific Contact Physical Model (CPM). (b) During inference, the CPM serves as an external model that steers the base policy’s action generation within the sampling process using a feasibility score. (c) In action space, TouchGuide can be viewed as a form of contact-physics steering that steers the policy distribution toward the real distri… view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of policy distributions in action space. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Five experiment tasks including Shoe Lacing, Chip Handover, Cucumber Peeling, Vase Wiping, and Lock Opening. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Steering hyperparameter choice on Chip Handover task using π0.5 as the base policy. (a) Performance differences across guidance scale (with guidance step KTouchGuide = 0.3). (b) Performance differences across guidance steps (with guidance scale η = 10). Guidance Scale. As shown in [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Common baseline failure cases for Shoe Lacing, Chip Handover, Cucumber Peeling, Vase Wiping, and Lock Opening. [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Trajectory visualization comparing SLAM-based UMI and TacUMI. (a)-(c) show three randomly sampled trajectories collected with SLAM-based UMI, and (e)-(g) show three randomly sampled trajectories collected with TacUMI. (d) and (h) overlay the trajectories from (a)-(c) and (e)-(g), respectively, illustrating trajectory consistency for the same task. Xense Tactile Sensor USB (Wrist) Camera RK3576 Core Board G… view at source ↗
Figure 9
Figure 9. Figure 9: TacUMI Hardware Design. (a) Collection-side: TCP pose is directly provided; a Vive Tracker and a magnetic encoder measure end-effector pose and gripper position, respectively. (b) Execution-side: a gripper motor actuates the gripper, with an identical mechanical structure for direct deployment. 1) Existing System Comparison: As shown in Table VIII, we compare our system with state-of-the-art (SOTA) data co… view at source ↗
Figure 10
Figure 10. Figure 10: The CPM feasibility score under in-distribution and out-of-distribution settings. In-distribution (Top): The dataset is [PITH_FULL_IMAGE:figures/full_fig_p023_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Feasibility score visualization. (a)–(c) We randomly sample from the dataset. The green box indicates the base frame we selected; we then expand a temporal window around it at 10 Hz, taking six frames before and six frames after, and visualize the corresponding feasibility scores (blue —) [PITH_FULL_IMAGE:figures/full_fig_p024_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Evaluation metrics for the Cucumber Peeling and Vase Wiping tasks. [PITH_FULL_IMAGE:figures/full_fig_p026_12.png] view at source ↗
read the original abstract

Fine-grained and contact-rich manipulation remain challenging for robots, largely due to the underutilization of tactile feedback. To address this, we introduce TouchGuide, a novel cross-policy visuo-tactile fusion paradigm that fuses modalities within a low-dimensional action space. Specifically, TouchGuide operates in two stages to guide a pre-trained diffusion or flow-matching visuomotor policy at inference time. First, the policy produces a coarse, visually-plausible action using only visual inputs during early sampling. Second, a task-specific Contact Physical Model (CPM) provides tactile guidance to steer and refine the action, ensuring it aligns with realistic physical contact conditions. Trained through contrastive learning on limited expert demonstrations, the CPM provides a tactile-informed feasibility score to steer the sampling process toward refined actions that satisfy physical contact constraints. Furthermore, to facilitate TouchGuide training with high-quality and cost-effective data, we introduce TacUMI, a data collection system. TacUMI achieves a favorable trade-off between precision and affordability; by leveraging rigid fingertips, it obtains direct tactile feedback, thereby enabling the collection of reliable tactile data. Extensive experiments on five challenging contact-rich tasks, such as shoe lacing and chip handover, show that TouchGuide consistently and significantly outperforms state-of-the-art visuo-tactile policies.

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

2 major / 0 minor

Summary. The paper introduces TouchGuide, a two-stage inference-time method for steering pre-trained diffusion or flow-matching visuomotor policies using tactile feedback. A task-specific Contact Physical Model (CPM), trained via contrastive learning on limited expert demonstrations, provides feasibility scores to refine coarse visual actions toward physically valid contacts. The work also presents TacUMI, a rigid-fingertip data collection system for affordable tactile data. Experiments on five contact-rich tasks (e.g., shoe lacing, chip handover) claim consistent and significant outperformance over state-of-the-art visuo-tactile policies.

Significance. If the quantitative results and generalization claims hold under scrutiny, TouchGuide offers a practical advance for contact-rich manipulation by enabling tactile-informed refinement at inference time without retraining the base policy. This could reduce the data and compute burden for fine-grained tasks while improving physical feasibility, provided the CPM proves robust beyond the training distribution.

major comments (2)
  1. Abstract: The claim of 'consistent and significant outperformance' on five tasks is stated without any quantitative metrics, baselines, error bars, or ablation details, which prevents assessment of effect sizes or statistical reliability.
  2. CPM training and evaluation sections: No ablations are reported on the number of expert demonstrations required for CPM training, cross-task transfer performance, or robustness to contact variations (friction, compliance, sensor noise). This leaves open the possibility that reported gains arise from overfitting to demonstration-specific contact patterns rather than a general steering mechanism.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback, which highlights opportunities to improve the clarity and robustness of our claims. We address each major comment below and will make targeted revisions to the manuscript.

read point-by-point responses

  1. Referee: Abstract: The claim of 'consistent and significant outperformance' on five tasks is stated without any quantitative metrics, baselines, error bars, or ablation details, which prevents assessment of effect sizes or statistical reliability.

    Authors: We agree that the abstract would benefit from including key quantitative results. The main paper (Section 5, Tables 1-3) already reports success rates with error bars, baselines, and statistical comparisons across all five tasks. In the revised version, we will update the abstract to preview representative metrics (e.g., average success rate improvements of 18-32% over state-of-the-art visuo-tactile baselines) while directing readers to the full experimental details. This change will allow better assessment of effect sizes without altering the underlying claims. revision: yes

  2. Referee: CPM training and evaluation sections: No ablations are reported on the number of expert demonstrations required for CPM training, cross-task transfer performance, or robustness to contact variations (friction, compliance, sensor noise). This leaves open the possibility that reported gains arise from overfitting to demonstration-specific contact patterns rather than a general steering mechanism.

    Authors: We appreciate this important point regarding potential overfitting. The current manuscript demonstrates TouchGuide's effectiveness using the collected demonstrations but does not include the requested ablations. In the revision, we will add: (i) an ablation study varying the number of expert demonstrations (5, 10, 20) for CPM training, (ii) cross-task transfer experiments where a CPM trained on one task is applied to another, and (iii) robustness tests introducing sensor noise and friction/compliance variations in simulation. These results will be included in the main text or supplementary material to support that the steering mechanism generalizes beyond demonstration-specific patterns. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical method with separate training and inference stages

full rationale

The paper describes TouchGuide as a two-stage inference-time steering procedure: a pre-trained visuomotor policy generates coarse actions from vision, then a separately trained task-specific Contact Physical Model (CPM) supplies feasibility scores to refine sampling. The CPM is trained contrastively on expert demonstrations and applied at test time; the headline results are experimental outperformance on five tasks. No equations, derivations, or self-citations are presented that reduce the claimed gains to the training inputs by construction. The generalization of the CPM is an empirical assumption, not a definitional or fitted-input circularity.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 2 invented entities

Central claim rests on the generalization of the CPM from limited demonstrations and the assumption that visual policies produce sufficiently coarse but plausible actions for subsequent tactile correction.

free parameters (1)
  • CPM contrastive learning hyperparameters
    Parameters for training the feasibility scorer on expert data are not specified.
axioms (1)
  • domain assumption Pre-trained diffusion or flow-matching visuomotor policies produce coarse visually-plausible actions in early sampling steps
    Invoked in the two-stage inference procedure described in the abstract.
invented entities (2)
  • Contact Physical Model (CPM) no independent evidence
    purpose: Supplies tactile-informed feasibility score to steer action sampling toward realistic contact conditions
    New model introduced and trained via contrastive learning on limited demonstrations.
  • TacUMI no independent evidence
    purpose: Data collection system that obtains direct tactile feedback using rigid fingertips
    Introduced to enable high-quality, cost-effective tactile data acquisition.

pith-pipeline@v0.9.0 · 5578 in / 1243 out tokens · 27221 ms · 2026-05-16T11:03:33.050569+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 4 Pith papers

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

  1. Referring-Aware Visuomotor Policy Learning for Closed-Loop Manipulation

    cs.RO 2026-04 unverdicted novelty 7.0

    ReV is a referring-aware visuomotor policy using coupled diffusion heads for real-time trajectory replanning in robotic manipulation, trained solely via targeted perturbations to expert demonstrations and achieving hi...

  2. Learning Tactile-Aware Quadrupedal Loco-Manipulation Policies

    cs.RO 2026-04 unverdicted novelty 6.0

    A tactile-aware hierarchical policy for quadrupedal loco-manipulation improves real-world contact-rich task performance by 28.54% over vision-only and visuotactile baselines.

  3. TAMEn: Tactile-Aware Manipulation Engine for Closed-Loop Data Collection in Contact-Rich Tasks

    cs.RO 2026-04 unverdicted novelty 6.0

    TAMEn supplies a cross-morphology wearable interface and pyramid-structured visuo-tactile data regime that raises bimanual manipulation success rates from 34% to 75% via closed-loop collection.

  4. Learning Tactile-Aware Quadrupedal Loco-Manipulation Policies

    cs.RO 2026-04 unverdicted novelty 5.0

    A hierarchical tactile-aware policy combines human-demonstration training for contact cue prediction with sim-to-real reinforcement learning to improve quadrupedal loco-manipulation performance by 28.54% over vision b...

Reference graph

Works this paper leans on

98 extracted references · 98 canonical work pages · cited by 3 Pith papers · 11 internal anchors

  1. [1]

    Aloha 2: An enhanced low-cost hardware for bimanual teleoperation,

    Jorge Aldaco, Travis Armstrong, Robert Baruch, Jeff Bingham, Sanky Chan, Kenneth Draper, De- bidatta Dwibedi, Chelsea Finn, Pete Florence, Spencer Goodrich, et al. Aloha 2: An enhanced low-cost hardware for bimanual teleoperation.arXiv preprint arXiv:2405.02292, 2024. 2

    work page arXiv 2024

  2. [2]

    https://www.arx-x.com/?product/, 2025

    ARX5. https://www.arx-x.com/?product/, 2025. 6

    work page 2025

  3. [3]

    Bifold: Bimanual cloth folding with language guidance.arXiv preprint arXiv:2501.16458, 2025

    Oriol Barbany, Adri `a Colom´e, and Carme Torras. Bifold: Bimanual cloth folding with language guidance.arXiv preprint arXiv:2501.16458, 2025. 1

    work page arXiv 2025

  4. [4]

    Vla-touch: Enhancing vision-language- action models with dual-level tactile feedback.arXiv preprint arXiv:2507.17294, 2025

    Jianxin Bi, Kevin Yuchen Ma, Ce Hao, Mike Zheng Shou, and Harold Soh. Vla-touch: Enhancing vision- language-action models with dual-level tactile feedback. arXiv preprint arXiv:2507.17294, 2025. 1, 2, 6

    work page arXiv 2025

  5. [5]

    $\pi_0$: A Vision-Language-Action Flow Model for General Robot Control

    Kevin Black, Noah Brown, Danny Driess, Adnan Es- mail, Michael Equi, Chelsea Finn, Niccolo Fusai, Lachy Groom, Karol Hausman, Brian Ichter, Szymon Jakubczak, Tim Jones, Liyiming Ke, Sergey Levine, Adrian Li-Bell, Mohith Mothukuri, Suraj Nair, Karl Pertsch, Lucy Xiaoyang Shi, James Tanner, Quan Vuong, Anna Walling, Haohuan Wang, and Ury Zhilinsky.π 0: A vi...

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  6. [6]

    Bi-act: Bilateral control-based imitation learning via action chunking with transformer

    Thanpimon Buamanee, Masato Kobayashi, Yuki Uran- ishi, and Haruo Takemura. Bi-act: Bilateral control-based imitation learning via action chunking with transformer. In2024 IEEE International Conference on Advanced Intelligent Mechatronics (AIM), pages 410–415. IEEE,

    work page

  7. [7]

    Lerobot: State-of-the-art machine learning for real-world robotics in pytorch

    Remi Cadene, Simon Alibert, Alexander Soare, Quentin Gallouedec, Adil Zouitine, Steven Palma, Pepijn Kooijmans, Michel Aractingi, Mustafa Shukor, Dana Aubakirova, Martino Russi, Francesco Capuano, Caro- line Pascal, Jade Choghari, Jess Moss, and Thomas Wolf. Lerobot: State-of-the-art machine learning for real-world robotics in pytorch. https://github.com/...

    work page 2024

  8. [8]

    Compose Your Policies! Improving Diffusion-based or Flow-based Robot Policies via Test-Time Distribution-Level Composition.arXiv preprint arXiv:2510.01068, 2025

    Jiahang Cao, Yize Huang, Hanzhong Guo, Rui Zhang, Mu Nan, Weijian Mai, Jiaxu Wang, Hao Cheng, Jingkai Sun, Gang Han, et al. Compose your policies! im- proving diffusion-based or flow-based robot policies via test-time distribution-level composition.arXiv preprint arXiv:2510.01068, 2025. 1, 6

    work page arXiv 2025

  9. [9]

    Multi-Modal Manipulation via Multi-Modal Policy Consensus

    Haonan Chen, Jiaming Xu, Hongyu Chen, Kaiwen Hong, Binghao Huang, Chaoqi Liu, Jiayuan Mao, Yunzhu Li, Yilun Du, and Katherine Driggs-Campbell. Multi-modal manipulation via multi-modal policy consensus.arXiv preprint arXiv:2509.23468, 2025. 1, 2, 6, 7, 17, 25, 28

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  10. [10]

    RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulation

    Tianxing Chen, Zanxin Chen, Baijun Chen, Zijian Cai, Yibin Liu, Zixuan Li, Qiwei Liang, Xianliang Lin, Yi- heng Ge, Zhenyu Gu, et al. Robotwin 2.0: A scalable data generator and benchmark with strong domain random- ization for robust bimanual robotic manipulation.arXiv preprint arXiv:2506.18088, 2025. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  11. [11]

    A simple framework for contrastive learning of visual representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations. InInternational con- ference on machine learning, pages 1597–1607. PmLR,

    work page

  12. [12]

    Visuo-tactile transformers for manipulation,

    Yizhou Chen, Andrea Sipos, Mark Van der Merwe, and Nima Fazeli. Visuo-tactile transformers for manipulation. arXiv preprint arXiv:2210.00121, 2022. 2

    work page arXiv 2022

  13. [13]

    Omnivtla: Vision- tactile-language-action model with semantic-aligned tactile sensing.arXiv preprint arXiv:2508.08706, 2025

    Zhengxue Cheng, Yiqian Zhang, Wenkang Zhang, Haoyu Li, Keyu Wang, Li Song, and Hengdi Zhang. Omnivtla: Vision-tactile-language-action model with semantic- aligned tactile sensing.arXiv preprint arXiv:2508.08706,

    work page arXiv

  14. [14]

    Universal Manipulation Interface: In-The-Wild Robot Teaching Without In-The-Wild Robots

    Cheng Chi, Zhenjia Xu, Chuer Pan, Eric Cousineau, Benjamin Burchfiel, Siyuan Feng, Russ Tedrake, and Shuran Song. Universal manipulation interface: In-the- wild robot teaching without in-the-wild robots.arXiv preprint arXiv:2402.10329, 2024. 2, 3, 4, 8, 18, 20, 28

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  15. [15]

    Diffusion policy: Visuomotor policy learning via action diffusion.The International Journal of Robotics Research, 44(10-11):1684–1704, 2025

    Cheng Chi, Zhenjia Xu, Siyuan Feng, Eric Cousineau, Yilun Du, Benjamin Burchfiel, Russ Tedrake, and Shuran Song. Diffusion policy: Visuomotor policy learning via action diffusion.The International Journal of Robotics Research, 44(10-11):1684–1704, 2025. 1, 2, 3, 4, 6, 18, 25, 28

    work page 2025

  16. [16]

    In-the-wild compliant manipulation with umi-ft.arXiv preprint arXiv:2601.09988, 2026

    Hojung Choi, Yifan Hou, Chuer Pan, Seongheon Hong, Austin Patel, Xiaomeng Xu, Mark R Cutkosky, and Shuran Song. In-the-wild compliant manipulation with umi-ft.arXiv preprint arXiv:2601.09988, 2026. 2, 18, 20

    work page arXiv 2026

  17. [17]

    Multimodal visual-tactile representation learning through self-supervised contrastive pre-training

    Vedant Dave, Fotios Lygerakis, and Elmar Rueckert. Multimodal visual-tactile representation learning through self-supervised contrastive pre-training. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 8013–8020. IEEE, 2024. 2

    work page 2024

  18. [18]

    Diffusion models beat gans on image synthesis.Advances in neural information processing systems, 34:8780–8794, 2021

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis.Advances in neural information processing systems, 34:8780–8794, 2021. 2, 4, 5, 14

    work page 2021

  19. [19]

    Using 3d mice to control robot manipulators

    Varad Dhat, Nick Walker, and Maya Cakmak. Using 3d mice to control robot manipulators. InProceedings of the 2024 ACM/IEEE International Conference on Human- Robot Interaction, pages 896–900, 2024. 2

    work page 2024

  20. [20]

    Bunny-visionpro: Real-time bimanual dexterous teleop- eration for imitation learning

    Runyu Ding, Yuzhe Qin, Jiyue Zhu, Chengzhe Jia, Shiqi Yang, Ruihan Yang, Xiaojuan Qi, and Xiaolong Wang. Bunny-visionpro: Real-time bimanual dexterous teleop- eration for imitation learning. In2025 IEEE/RSJ Inter- national Conference on Intelligent Robots and Systems (IROS), pages 12248–12255. IEEE, 2025. 2, 18, 19

    work page 2025

  21. [21]

    Adaptive visual–tactile fusion recognition for robotic operation of multi-material system.Frontiers in Neurorobotics, 17:1181383, 2023

    Zihao Ding, Guodong Chen, Zhenhua Wang, and Lin- ing Sun. Adaptive visual–tactile fusion recognition for robotic operation of multi-material system.Frontiers in Neurorobotics, 17:1181383, 2023. 2

    work page 2023

  22. [22]

    Chung, H., Kim, J., McCann, M

    Maximilian Du and Shuran Song. Dynaguide: Steering diffusion polices with active dynamic guidance.arXiv preprint arXiv:2506.13922, 2025. 2, 3, 4, 5, 22, 25, 28

    work page arXiv 2025

  23. [23]

    On the guidance of flow matching.arXiv preprint arXiv:2502.02150, 2025

    Ruiqi Feng, Chenglei Yu, Wenhao Deng, Peiyan Hu, and Tailin Wu. On the guidance of flow matching.arXiv preprint arXiv:2502.02150, 2025. 4, 5, 14

    work page arXiv 2025

  24. [24]

    https://www.flexiv.com/products/rizon,

    Flexiv Rizon4. https://www.flexiv.com/products/rizon,

    work page

  25. [25]

    Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperation

    Zipeng Fu, Tony Z Zhao, and Chelsea Finn. Mo- bile aloha: Learning bimanual mobile manipulation with low-cost whole-body teleoperation.arXiv preprint arXiv:2401.02117, 2024. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  26. [26]

    Vital pretraining: Visuo-tactile pretraining for tactile and non-tactile manipulation poli- cies

    Abraham George, Selam Gano, Pranav Katragadda, and Amir Barati Farimani. Vital pretraining: Visuo-tactile pretraining for tactile and non-tactile manipulation poli- cies. In2025 IEEE International Conference on Robotics and Automation (ICRA), pages 258–264. IEEE, 2025. 2

    work page 2025

  27. [27]

    Tla: Tactile- language-action model for contact-rich manipulation.arXiv preprint arXiv:2503.08548, 2025

    Peng Hao, Chaofan Zhang, Dingzhe Li, Xiaoge Cao, Xiaoshuai Hao, Shaowei Cui, and Shuo Wang. Tla: Tactile-language-action model for contact-rich manipu- lation.arXiv preprint arXiv:2503.08548, 2025. 2

    work page arXiv 2025

  28. [28]

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

    Erik Helmut, Niklas Funk, Tim Schneider, Cristiana de Farias, and Jan Peters. Tactile-conditioned diffu- sion policy for force-aware robotic manipulation.arXiv preprint arXiv:2510.13324, 2025. 2, 18, 20

    work page arXiv 2025

  29. [29]

    Huang, Y

    Binghao Huang, Yixuan Wang, Xinyi Yang, Yiyue Luo, and Yunzhu Li. 3d-vitac: Learning fine-grained ma- nipulation with visuo-tactile sensing.arXiv preprint arXiv:2410.24091, 2024. 2

    work page arXiv 2024

  30. [30]

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

    Jialei Huang, Shuo Wang, Fanqi Lin, Yihang Hu, Chuan Wen, and Yang Gao. Tactile-vla: unlocking vision- language-action model’s physical knowledge for tactile generalization.arXiv preprint arXiv:2507.09160, 2025. 2

    work page arXiv 2025

  31. [31]

    Physical Intelligence, Ali Amin, Raichelle Aniceto, Ash- win Balakrishna, Kevin Black, Ken Conley, Grace Con- nors, James Darpinian, Karan Dhabalia, Jared DiCarlo, et al.π ∗ 0.6: a vla that learns from experience.arXiv preprint arXiv:2511.14759, 2025. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  32. [32]

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

    Physical Intelligence, Kevin Black, Noah Brown, James Darpinian, Karan Dhabalia, Danny Driess, Adnan Es- mail, Michael Equi, Chelsea Finn, Niccolo Fusai, et al. π0.5: a vision-language-action model with open-world generalization.arXiv preprint arXiv:2504.16054, 2025. 2, 3, 4, 6, 15, 18, 25, 28

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  33. [33]

    OPEN TEACH: A versatile teleoperation system for robotic manipulation.arXiv preprint arXiv:2403.07870,

    Aadhithya Iyer, Zhuoran Peng, Yinlong Dai, Irmak Guzey, Siddhant Haldar, Soumith Chintala, and Lerrel Pinto. Open teach: A versatile teleoperation system for robotic manipulation.arXiv preprint arXiv:2403.07870,

    work page arXiv

  34. [34]

    Streaming flow policy: Simplifying diffusion/flow- matching policies by treating action trajectories as flow trajectories.arXiv preprint arXiv:2505.21851, 2025

    Sunshine Jiang, Xiaolin Fang, Nicholas Roy, Tom ´as Lozano-P´erez, Leslie Pack Kaelbling, and Siddharth An- cha. Streaming flow policy: Simplifying diffusion/flow- matching policies by treating action trajectories as flow trajectories.arXiv preprint arXiv:2505.21851, 2025. 2

    work page arXiv 2025

  35. [35]

    Learning variable compliance control from a few demonstrations for bimanual robot with haptic feedback teleoperation system

    Tatsuya Kamijo, Cristian C Beltran-Hernandez, and Masashi Hamaya. Learning variable compliance control from a few demonstrations for bimanual robot with haptic feedback teleoperation system. In2024 IEEE/RSJ Inter- national Conference on Intelligent Robots and Systems (IROS), pages 12663–12670. IEEE, 2024. 2

    work page 2024

  36. [36]

    Soft-bubble grippers for robust and perceptive manipu- lation

    Naveen Kuppuswamy, Alex Alspach, Avinash Uttam- chandani, Sam Creasey, Takuya Ikeda, and Russ Tedrake. Soft-bubble grippers for robust and perceptive manipu- lation. In2020 IEEE/RSJ international conference on intelligent robots and systems (IROS), pages 9917–9924. IEEE, 2020. 2

    work page 2020

  37. [37]

    Digit: A novel design for a low-cost compact high-resolution tactile sensor with application to in-hand manipulation.IEEE Robotics and Automation Letters, 5 (3):3838–3845, 2020

    Mike Lambeta, Po-Wei Chou, Stephen Tian, Brian Yang, Benjamin Maloon, Victoria Rose Most, Dave Stroud, Raymond Santos, Ahmad Byagowi, Gregg Kammerer, et al. Digit: A novel design for a low-cost compact high-resolution tactile sensor with application to in-hand manipulation.IEEE Robotics and Automation Letters, 5 (3):3838–3845, 2020. 2

    work page 2020

  38. [38]

    Making sense of vision and touch: Self-supervised learning of multimodal representations for contact-rich tasks

    Michelle A Lee, Yuke Zhu, Krishnan Srinivasan, Parth Shah, Silvio Savarese, Li Fei-Fei, Animesh Garg, and Jeannette Bohg. Making sense of vision and touch: Self-supervised learning of multimodal representations for contact-rich tasks. In2019 International conference on robotics and automation (ICRA), pages 8943–8950. IEEE, 2019. 2

    work page 2019

  39. [39]

    Vitamin-b: A reliable and efficient visuo- tactile bimanual manipulation interface.arXiv preprint arXiv:2511.05858, 2025

    Chuanyu Li, Chaoyi Liu, Daotan Wang, Shuyu Zhang, Lusong Li, Zecui Zeng, Fangchen Liu, Jing Xu, and Rui Chen. Vitamin-b: A reliable and efficient visuo- tactile bimanual manipulation interface.arXiv preprint arXiv:2511.05858, 2025. 2, 18, 20

    work page arXiv 2025

  40. [40]

    DTact: A Vision-Based Tactile Sensor that Measures High-Resolution 3D Geometry Directly from Darkness,

    Changyi Lin, Ziqi Lin, Shaoxiong Wang, and Huazhe Xu. Dtact: A vision-based tactile sensor that measures high- resolution 3d geometry directly from darkness.arXiv preprint arXiv:2209.13916, 2022. 2

    work page arXiv 2022

  41. [41]

    9dtact: A compact vision-based tactile sensor for accurate 3d shape reconstruction and generalizable 6d force estimation.IEEE Robotics and Automation Letters, 9(2):923–930, 2023

    Changyi Lin, Han Zhang, Jikai Xu, Lei Wu, and Huazhe Xu. 9dtact: A compact vision-based tactile sensor for accurate 3d shape reconstruction and generalizable 6d force estimation.IEEE Robotics and Automation Letters, 9(2):923–930, 2023. 2

    work page 2023

  42. [42]

    Vitamin: Learning contact- rich tasks through robot-free visuo-tactile manipulation interface.arXiv preprint arXiv:2504.06156, 2025

    Fangchen Liu, Chuanyu Li, Yihua Qin, Jing Xu, Pieter Abbeel, and Rui Chen. Vitamin: Learning contact- rich tasks through robot-free visuo-tactile manipulation interface.arXiv preprint arXiv:2504.06156, 2025. 2, 18, 20

    work page arXiv 2025

  43. [43]

    Fastumi-100k: Advancing data-driven robotic manipulation with a large-scale umi-style dataset.arXiv preprint arXiv:2510.08022, 2025

    Kehui Liu, Zhongjie Jia, Yang Li, Pengan Chen, Song Liu, Xin Liu, Pingrui Zhang, Haoming Song, Xinyi Ye, Nieqing Cao, et al. Fastumi-100k: Advancing data-driven robotic manipulation with a large-scale umi-style dataset. arXiv preprint arXiv:2510.08022, 2025. 2

    work page arXiv 2025

  44. [44]

    CDP: Towards Robust Autoregressive Visuomotor Policy Learning via Causal Diffusion, August 2025

    Jiahua Ma, Yiran Qin, Yixiong Li, Xuanqi Liao, Yulan Guo, and Ruimao Zhang. Cdp: Towards robust autore- gressive visuomotor policy learning via causal diffusion. arXiv preprint arXiv:2506.14769, 2025. 1, 2

    work page arXiv 2025

  45. [45]

    A live-stream robotic teamwork for clothing manipulation from zero to hero.HKU MMLab Research Blog, 2025

    HKU MMLab. A live-stream robotic teamwork for clothing manipulation from zero to hero.HKU MMLab Research Blog, 2025. https://mmlab.hk/research/kai0. 1

    work page 2025

  46. [46]

    Robotwin: Dual-arm robot benchmark with generative digital twins (early version)

    Yao Mu, Tianxing Chen, Shijia Peng, Zanxin Chen, Zeyu Gao, Yude Zou, Lunkai Lin, Zhiqiang Xie, and Ping Luo. Robotwin: Dual-arm robot benchmark with generative digital twins (early version). InEuropean Conference on Computer Vision, pages 264–273. Springer, 2024. 2

    work page 2024

  47. [47]

    Robofactory: Exploring embodied agent collab- oration with compositional constraints.arXiv preprint arXiv:2503.16408, 2025

    Yiran Qin, Li Kang, Xiufeng Song, Zhenfei Yin, Xiaohong Liu, Xihui Liu, Ruimao Zhang, and Lei Bai. Robofactory: Exploring embodied agent collab- oration with compositional constraints.arXiv preprint arXiv:2503.16408, 2025. 1, 25, 28

    work page arXiv 2025

  48. [48]

    Anyteleop: A general vision-based dexterous robot arm-hand teleoperation system.arXiv preprint arXiv:2307.04577, 2023

    Yuzhe Qin, Wei Yang, Binghao Huang, Karl Van Wyk, Hao Su, Xiaolong Wang, Yu-Wei Chao, and Dieter Fox. Anyteleop: A general vision-based dexterous robot arm-hand teleoperation system.arXiv preprint arXiv:2307.04577, 2023. 2

    work page arXiv 2023

  49. [49]

    Mc-tac: Mod- ular camera-based tactile sensor for robot gripper

    Jieji Ren, Jiang Zou, and Guoying Gu. Mc-tac: Mod- ular camera-based tactile sensor for robot gripper. In International Conference on Intelligent Robotics and Applications, pages 169–179. Springer, 2023. 2

    work page 2023

  50. [50]

    SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics

    Mustafa Shukor, Dana Aubakirova, Francesco Ca- puano, Pepijn Kooijmans, Steven Palma, Adil Zouitine, Michel Aractingi, Caroline Pascal, Martino Russi, An- dres Marafioti, et al. Smolvla: A vision-language-action model for affordable and efficient robotics.arXiv preprint arXiv:2506.01844, 2025. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  51. [51]

    Latent policy bar- rier: Learning robust visuomotor policies by staying in- distribution.arXiv preprint arXiv:2508.05941, 2025

    Zhanyi Sun and Shuran Song. Latent policy bar- rier: Learning robust visuomotor policies by staying in- distribution.arXiv preprint arXiv:2508.05941, 2025. 2, 3, 4, 5, 6, 25, 28

    work page arXiv 2025

  52. [52]

    Neuralfeels with neural fields: Visuotactile perception for in-hand manipulation.Science Robotics, 9(96):eadl0628,

    Sudharshan Suresh, Haozhi Qi, Tingfan Wu, Taosha Fan, Luis Pineda, Mike Lambeta, Jitendra Malik, Mrinal Kalakrishnan, Roberto Calandra, Michael Kaess, et al. Neuralfeels with neural fields: Visuotactile perception for in-hand manipulation.Science Robotics, 9(96):eadl0628,

    work page

  53. [53]

    Gel- slim 3.0: High-resolution measurement of shape, force and slip in a compact tactile-sensing finger

    Ian H Taylor, Siyuan Dong, and Alberto Rodriguez. Gel- slim 3.0: High-resolution measurement of shape, force and slip in a compact tactile-sensing finger. In2022 International Conference on Robotics and Automation (ICRA), pages 10781–10787. IEEE, 2022. 2

    work page 2022

  54. [54]

    https://www.vive.com/hk/accessory/ tracker3, 2025

    Vive Tracker. https://www.vive.com/hk/accessory/ tracker3, 2025. 2

    work page 2025

  55. [55]

    Inference-time policy steering through human interactions

    Yanwei Wang, Lirui Wang, Yilun Du, Balakumar Sun- daralingam, Xuning Yang, Yu-Wei Chao, Claudia P ´erez- D’Arpino, Dieter Fox, and Julie Shah. Inference-time policy steering through human interactions. In2025 IEEE International Conference on Robotics and Automa- tion (ICRA), pages 15626–15633. IEEE, 2025. 2, 4

    work page 2025

  56. [56]

    Gaudp: Rein- venting multi-agent collaboration through gaussian- image synergy in diffusion policies.arXiv preprint arXiv:2511.00998, 2025

    Ziye Wang, Li Kang, Yiran Qin, Jiahua Ma, Zhanglin Peng, Lei Bai, and Ruimao Zhang. Gaudp: Rein- venting multi-agent collaboration through gaussian- image synergy in diffusion policies.arXiv preprint arXiv:2511.00998, 2025. 1, 2

    work page arXiv 2025

  57. [57]

    Ensuring force safety in vision-guided robotic manip- ulation via implicit tactile calibration.arXiv preprint arXiv:2412.10349, 2024

    Lai Wei, Jiahua Ma, Yibo Hu, and Ruimao Zhang. Ensuring force safety in vision-guided robotic manip- ulation via implicit tactile calibration.arXiv preprint arXiv:2412.10349, 2024. 1, 6, 17

    work page arXiv 2024

  58. [58]

    Diffusion-vla: Scal- ing robot foundation models via unified diffusion and autoregression.arXiv preprint arXiv:2412.03293, 2024

    Junjie Wen, Minjie Zhu, Yichen Zhu, Zhibin Tang, Jinming Li, Zhongyi Zhou, Chengmeng Li, Xiaoyu Liu, Yaxin Peng, Chaomin Shen, et al. Diffusion-vla: General- izable and interpretable robot foundation model via self- generated reasoning.arXiv preprint arXiv:2412.03293,

    work page arXiv

  59. [59]

    Freetacman: Robot-free visuo-tactile data col- lection system for contact-rich manipulation,

    Longyan Wu, Checheng Yu, Jieji Ren, Li Chen, Yufei Jiang, Ran Huang, Guoying Gu, and Hongyang Li. Freetacman: Robot-free visuo-tactile data collection sys- tem for contact-rich manipulation.arXiv preprint arXiv:2506.01941, 2025. 2, 18, 20

    work page arXiv 2025

  60. [60]

    Gello: A general, low-cost, and intuitive teleoperation framework for robot manipulators

    Philipp Wu, Yide Shentu, Zhongke Yi, Xingyu Lin, and Pieter Abbeel. Gello: A general, low-cost, and intuitive teleoperation framework for robot manipulators. In2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 12156–12163. IEEE, 2024. 2, 18, 19

    work page 2024

  61. [61]

    https://www.xenserobotics.com/product/367/ detail/9, 2025

    Xense. https://www.xenserobotics.com/product/367/ detail/9, 2025. 2

    work page 2025

  62. [62]

    https://xensedoc.readthedocs.io/en/latest/ XenseSDK/XenseSDK.html, 2025

    Xense SDK. https://xensedoc.readthedocs.io/en/latest/ XenseSDK/XenseSDK.html, 2025. 2, 3, 25

    work page 2025

  63. [63]

    Dexumi: Using human hand as the universal manipulation in- terface for dexterous manipulation.arXiv preprint arXiv:2505.21864, 2025

    Mengda Xu, Han Zhang, Yifan Hou, Zhenjia Xu, Linxi Fan, Manuela Veloso, and Shuran Song. Dexumi: Using human hand as the universal manipulation in- terface for dexterous manipulation.arXiv preprint arXiv:2505.21864, 2025. 2

    work page arXiv 2025

  64. [64]

    exumi: Extensible robot teaching system with action-aware task-agnostic tactile representation

    Yue Xu, Litao Wei, Pengyu An, Qingyu Zhang, and Yong-Lu Li. exumi: Extensible robot teaching system with action-aware task-agnostic tactile representation. arXiv preprint arXiv:2509.14688, 2025. 2, 18, 20

    work page arXiv 2025

  65. [65]

    Reactive diffusion policy: Slow-fast visual-tactile policy learning for contact-rich manipulation,

    Han Xue, Jieji Ren, Wendi Chen, Gu Zhang, Yuan Fang, Guoying Gu, Huazhe Xu, and Cewu Lu. Re- active diffusion policy: Slow-fast visual-tactile policy learning for contact-rich manipulation.arXiv preprint arXiv:2503.02881, 2025. 1, 2, 3, 6, 7, 17, 18, 19, 25, 28

    work page arXiv 2025

  66. [66]

    Touch and go: Learning from human-collected vision and touch,

    Fengyu Yang, Chenyang Ma, Jiacheng Zhang, Jing Zhu, Wenzhen Yuan, and Andrew Owens. Touch and go: Learning from human-collected vision and touch.arXiv preprint arXiv:2211.12498, 2022. 2

    work page arXiv 2022

  67. [67]

    Demonstrating the octopi- 1.5 visual-tactile-language model,

    Samson Yu, Kelvin Lin, and Harold Soh. Demonstrat- ing the octopi-1.5 visual-tactile-language model.arXiv preprint arXiv:2507.09985, 2025. 2

    work page arXiv 2025

  68. [68]

    Gelsight: High-resolution robot tactile sensors for esti- mating geometry and force.Sensors, 17(12):2762, 2017

    Wenzhen Yuan, Siyuan Dong, and Edward H Adelson. Gelsight: High-resolution robot tactile sensors for esti- mating geometry and force.Sensors, 17(12):2762, 2017. 2

    work page 2017

  69. [69]

    3D Diffusion Policy: Generalizable Visuomotor Policy Learning via Simple 3D Representations

    Yanjie Ze, Gu Zhang, Kangning Zhang, Chenyuan Hu, Muhan Wang, and Huazhe Xu. 3d diffusion policy: Generalizable visuomotor policy learning via simple 3d representations.arXiv preprint arXiv:2403.03954, 2024. 1, 2

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  70. [70]

    Vtla: Vision- tactile-language-action model with preference learning for insertion manipulation,

    Chaofan Zhang, Peng Hao, Xiaoge Cao, Xiaoshuai Hao, Shaowei Cui, and Shuo Wang. Vtla: Vision-tactile- language-action model with preference learning for in- sertion manipulation.arXiv preprint arXiv:2505.09577,

    work page arXiv

  71. [71]

    Flowpolicy: Enabling fast and robust 3d flow-based policy via consistency flow matching for robot manipulation

    Qinglun Zhang, Zhen Liu, Haoqiang Fan, Guanghui Liu, Bing Zeng, and Shuaicheng Liu. Flowpolicy: Enabling fast and robust 3d flow-based policy via consistency flow matching for robot manipulation. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 14754–14762, 2025. 2

    work page 2025

  72. [72]

    arXiv preprint arXiv:2505.23614 , year =

    Xiangcheng Zhang, Haowei Lin, Haotian Ye, James Zou, Jianzhu Ma, Yitao Liang, and Yilun Du. Inference- time scaling of diffusion models through classical search. arXiv preprint arXiv:2505.23614, 2025. 2

    work page arXiv 2025

  73. [73]

    Effective estimation of contact force and torque for vision-based tactile sensors with helmholtz–hodge decomposition.IEEE Robotics and Automation Letters, 4(4):4094–4101, 2019

    Yazhan Zhang, Zicheng Kan, Yang Yang, Yu Alexander Tse, and Michael Yu Wang. Effective estimation of contact force and torque for vision-based tactile sensors with helmholtz–hodge decomposition.IEEE Robotics and Automation Letters, 4(4):4094–4101, 2019. 2

    work page 2019

  74. [74]

    ifem2.0: Dense 3d contact force field reconstruction and assessment for vision-based tactile sensors.IEEE Transactions on Robotics, 2024

    Can Zhao, Jin Liu, and Daolin Ma. ifem2.0: Dense 3d contact force field reconstruction and assessment for vision-based tactile sensors.IEEE Transactions on Robotics, 2024. 2

    work page 2024

  75. [75]

    Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware

    Tony Z Zhao, Vikash Kumar, Sergey Levine, and Chelsea Finn. Learning fine-grained bimanual manipulation with low-cost hardware.arXiv preprint arXiv:2304.13705,

    work page internal anchor Pith review Pith/arXiv arXiv

  76. [76]

    Fastumi: A scalable and hardware-independent universal manip- ulation interface with dataset

    Zhaxizhuom Zhaxizhuoma, Kehui Liu, Chuyue Guan, Zhongjie Jia, Ziniu Wu, Xin Liu, Tianyu Wang, Shuai Liang, Pengan CHEN, Pingrui Zhang, et al. Fastumi: A scalable and hardware-independent universal manip- ulation interface with dataset. InConference on Robot Learning, pages 3069–3093. PMLR, 2025. 2, 3, 18, 20, 28

    work page 2025

  77. [77]

    Guided flows for generative modeling and decision making.arXiv preprint arXiv:2311.13443, 2023

    Qinqing Zheng, Matt Le, Neta Shaul, Yaron Lipman, Aditya Grover, and Ricky TQ Chen. Guided flows for generative modeling and decision making.arXiv preprint arXiv:2311.13443, 2023. 4, 5, 14

    work page arXiv 2023

  78. [78]

    DINO-WM: World Models on Pre-trained Visual Features enable Zero-shot Planning

    Gaoyue Zhou, Hengkai Pan, Yann LeCun, and Lerrel Pinto. Dino-wm: World models on pre-trained vi- sual features enable zero-shot planning.arXiv preprint arXiv:2411.04983, 2024. 4

    work page internal anchor Pith review arXiv 2024

  79. [79]

    Touch in the wild: Learning fine-grained manipulation with a portable visuo-tactile gripper.arXiv preprint arXiv:2507.15062, 2025

    Xinyue Zhu, Binghao Huang, and Yunzhu Li. Touch in the wild: Learning fine-grained manipulation with a portable visuo-tactile gripper.arXiv preprint arXiv:2507.15062, 2025. 2, 18, 20 APPENDIX A Classifier Guidance for Flow Matching (Proof of Proposition 1) . . . . . . . . . . . . . . . . . . . . . . 14 B Steering Hyperparameter Investigation . . . . . . ....

    work page arXiv 2025

  80. [80]

    We primarily varied two hyperparameters (i.e., guidance scaleηand guidance stepsK TouchGuide, for the detailed hyperparameter implementation, see Alg

    Ablation Study on Steering Hyperparameter:To select TouchGuide steering hyperparameters, we conducted extensive experiments on the Chip Handover task usingπ 0.5 [32] as the base policy. We primarily varied two hyperparameters (i.e., guidance scaleηand guidance stepsK TouchGuide, for the detailed hyperparameter implementation, see Alg. 1) by sweeping one w...

    work page

Showing first 80 references.