Pith. sign in

REVIEW 3 major objections 6 minor 34 references

Contact-rich robot policies can keep many approach paths before contact and still react fast to force once contact begins by switching sampling frequency mid-episode.

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

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

T0 review · grok-4.5

2026-07-31 02:43 UTC pith:DQGJ33R3

load-bearing objection Solid systems fix for a real contact-rich tradeoff; gains look real, but the multimodality gate is under-isolated. the 3 major comments →

arxiv 2607.28596 v1 pith:DQGJ33R3 submitted 2026-07-30 cs.RO

FA-RDP: A Frequency-Adaptive Reactive Diffusion Policy for Contact-Rich Manipulation

Lifeng Zhuo , Wendi Chen , Han Xue , Shirun Tang , Jun Lv , Cewu Lu , Chuan Wen This is my paper
classification cs.RO
keywords contact-rich manipulationdiffusion policyvisual-force controlfrequency-adaptive inferencemultimodalityconsistency distillationclosed-loop force feedback
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.

Standard diffusion robot policies pick one inference speed and one sampling depth for a whole trial. That forces a bad trade: slow multi-step sampling keeps several valid approach paths but is sluggish once force feedback matters; fast sampling reacts to wrench but tends to collapse those paths into one mode. This paper claims the two needs dominate different phases of the same episode, so the policy should change frequency when action ambiguity drops. FA-RDP uses one shared visual-force network that can emit both low- and high-frequency action chunks, a learned multimodality score from vision that chooses multi-step 10 Hz sampling before contact and one-step 30 Hz sampling after, and a distillation that teaches the fast sampler to predict actions on the robot’s action manifold rather than noise-like targets. On three real contact-rich tasks—box flipping, switch toggling, and button pressing—it reports the highest average success while still covering the diverse pre-contact approach modes.

Core claim

FA-RDP shows that phase-adaptive inference frequency resolves the multimodality–reactivity tradeoff in end-to-end visual-force diffusion policies: multi-step low-frequency sampling before contact plus one-step high-frequency manifold-distilled sampling after contact yields 81.7% average success on three contact-rich tasks—about 30 points above the strongest fixed-frequency end-to-end baseline—while preserving diverse pre-contact trajectory modes.

What carries the argument

Manifold Consistency Distillation (MCD) plus a multimodality indicator: MCD reparameterizes the high-frequency diffusion network to predict action chunks on the robot action manifold (with residual DDPM supervision and sample regression), enabling stable one-step closed-loop force response; the indicator, trained from low-frequency action residual scatter under fixed conditions, thresholds which sampler runs.

Load-bearing premise

A single learned score of how spread-out low-frequency action samples are, cut at a fixed threshold, is enough to mark when the robot should leave diverse slow planning and switch to fast force reaction across tasks.

What would settle it

On the same three tasks, freeze or reverse the indicator switch (always high-frequency distilled, always low-frequency multi-step, or switch at the wrong force/time) and check whether average success and the four-mode pre-contact approach distribution both collapse relative to the reported adaptive policy.

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

If this is right

  • Contact-rich visuomotor diffusion need not fix one control rate for an entire episode; rate can track estimated action ambiguity.
  • One shared backbone with frequency-aware positional encoding can serve both multimodal approach and dense force correction without separate slow/fast models.
  • Predicting actions on the robot action manifold, rather than epsilon/score/velocity, is a workable path to one-step high-rate closed-loop diffusion control.
  • Preserving multi-step diffusion only while pre-contact modes remain open can raise success without giving up force reactivity after contact.

Where Pith is reading between the lines

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

  • If the indicator is mostly tracking contact onset via visual cues correlated with force, simpler contact or wrench thresholds might recover much of the gain with less training.
  • The same phase split—multimodal open-loop-ish approach, then high-rate residual force correction—could transfer to other contact skills (insertion, wiping, assembly) where pre-contact geometry is underconstrained.
  • Three-stage training (joint multi-frequency diffusion, indicator head, then MCD) is a practical cost; joint or online adaptation of the switch would be a natural next stress test.

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 / 6 minor

Summary. FA-RDP proposes a frequency-adaptive end-to-end visual-force diffusion policy for contact-rich manipulation. A shared multi-frequency Transformer with frequency-adaptive positional encoding predicts both 10 Hz and 30 Hz action chunks; a learned scalar multimodality indicator on slow visual tokens selects multi-step low-frequency DDIM before contact and a one-step high-frequency sampler after contact; and Manifold Consistency Distillation (MCD) reparameterizes the high-frequency network to predict clean actions on the robot action manifold (with DDPM residual conversion plus sample regression) for stable one-step inference. On three real dual-object contact tasks (box flip, switch toggle, button press), FA-RDP reports 81.7% average success versus 51.7% for ImplicitRDP and 61.7% for high-frequency distilled alone, while mode histograms show preserved pre-contact trajectory diversity relative to always-high-frequency distillation.

Significance. If the result holds, the paper offers a practical and well-motivated resolution of a real systems tradeoff in reactive diffusion policies: fixed multi-step sampling is too slow after contact, while fixed one-step/high-frequency sampling collapses pre-contact modes. The shared multi-frequency backbone, consistent closed-loop force refresh (Alg. 1), and MCD formulation (predict actions on the manifold rather than epsilon/score/velocity) are concrete engineering contributions with clear ablations against MeanFlow and Consistency Policy (Table III). Real-robot comparisons against DP, RDP, ImplicitRDP, and regression, plus public code/demos, make the work useful to the contact-rich imitation-learning community even if some causal claims need tightening.

major comments (3)
  1. [Sec. IV-B, Table II, Fig. 9] Sec. IV-B, Eqs. (5)–(6), Alg. 1, Fig. 9, Table II: The central claim attributes the jump from 61.7% (high-frequency distilled alone) to 81.7% to a residual-trained multimodality indicator that gates sampler choice. Fig. 9 shows the indicator rising in lockstep with measured force at contact on all three tasks, so a force-threshold, contact flag, or fixed time schedule with the same two samplers is a direct confounder. Without that control, Table II shows that switching helps, not that residual-based “multimodality” is the causal gating signal. Please add at least one oracle/phase baseline (force threshold, binary contact, or fixed pre/post schedule) using identical low- and high-frequency samplers, and report how τ was chosen (cross-validation vs. hand-set 3.5).
  2. [Sec. V-C, Tables I–III] Tables I–III: All success rates are n=20 trials per task with no confidence intervals, standard errors, or statistical tests. Several pairwise gaps that support the narrative (e.g., FA-RDP 14/20 vs. HF-distilled 12/20 on Box; ImplicitRDP 8/20 vs. FA-RDP 14/20) are small in absolute counts. For a systems claim of +30 pp over the strongest fixed end-to-end baseline and +20 pp from indicator switching, please report binomial CIs or bootstrap intervals and, where feasible, increase trial count or pool with a pre-registered evaluation protocol so the ranking is not sensitive to a few trials.
  3. [Sec. IV-C, Eq. (7), Table III] Sec. IV-C, Eq. (7) and Table III: MCD is a main contribution, and the large gap vs. MeanFlow/Consistency Policy (61.7% vs. 1.7%) is striking but under-analyzed. The comparison confounds target parameterization (action-manifold prediction + DDPM residual) with the full MCD+SRL recipe, EMA teacher, and six-step grid G. A minimal ablation—same backbone and one-step budget, swapping only the prediction target (epsilon/velocity vs. clean action) and/or removing SRL—would show whether “manifold prediction” is the load-bearing design choice claimed in the abstract and Sec. II-C, or whether other training details dominate.
minor comments (6)
  1. [Sec. IV-D] Sec. IV-D: The shared 100 Hz force compensation (Eq. 8, λ=10^{-4}) is applied to all methods and is appropriate for fairness, but its interaction with high-frequency closed-loop control should be stated more clearly—e.g., whether gains would shrink under a stiffer or uncompensated low-level controller.
  2. [Fig. 8, Sec. V-C.2] Fig. 8 mode counts: Clarify how the four approach modes (Right/Mid-R/Mid-L/Left) were labeled (manual annotation criteria, automatic clustering) and whether labeling was blinded to method.
  3. [Sec. IV-B, Eq. (6)] Eq. (6): σ(o_n)=1+exp(s(o_n)) is an unusual softplus-style map named like a sigmoid; a one-line note that this follows VGGT’s confidence parameterization would reduce confusion with a probability in [0,1].
  4. [Abstract, Sec. I] Abstract vs. body: abstract says “Code and videos”; intro says “Code and demos” at fa-rdp.github.io. Align wording and ensure the linked page includes training configs for the three stages and the threshold τ used at deployment.
  5. [Sec. II-B] Related work: briefly distinguish FA-RDP’s phase-driven frequency switch from DVAC’s denoising-variance adaptive replanning and ManipForce’s frequency-aware fusion so the novelty boundary is explicit for multi-frequency readers.
  6. Typographical consistency: “ImplicitRDP” spacing/capitalization and “force/torque” vs. “wrench” alternate; pick one convention. Also fix “Noematrix Ltd.” affiliation formatting if required by the venue.

Circularity Check

0 steps flagged

No significant circularity: empirical systems paper whose claims are external task success rates and mode counts, not algebraic restatements of fitted inputs.

full rationale

FA-RDP’s load-bearing claims are measured outcomes on three real contact-rich tasks (Table I–III success rates; Fig. 8 pre-contact mode histograms), not quantities forced by construction from the training objectives. The multi-frequency diffusion loss (Eqs. 1–4), multimodality indicator regression on low-frequency action residuals (Eqs. 5–6), and manifold consistency distillation (Eq. 7) are standard supervised/distillation training recipes; none redefine the reported success metric. Self-citations to RDP and ImplicitRDP supply the shared slow-fast visual-force backbone and consistent closed-loop inference (Sec. III, Alg. 1), which is normal incremental systems work—the paper then evaluates FA-RDP against those baselines in new trials rather than treating the citations as uniqueness theorems that forbid alternatives. Correlation of the indicator with force onset (Fig. 9) and the lack of an oracle force/schedule control are causal-identification and correctness concerns, not circular derivation: the +20 pp gain (Table II) remains an external empirical comparison between two deployed policies. No step reduces a claimed prediction to its fitted input by identity.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 3 invented entities

Load-bearing content is mostly engineering assumptions from diffusion IL and the authors’ prior reactive diffusion stack, plus several hand-set control/distillation knobs. No new physical law is postulated; invented pieces are architectural modules whose only evidence is this paper’s tasks.

free parameters (6)
  • multimodality indicator threshold τ = 3.5 (figure); general τ in Alg. 1
    Fixed threshold selects low- vs high-frequency sampler at inference; Fig. 9 marks threshold=3.5 without a selection procedure or sensitivity sweep that the central gain depends on.
  • force compensation gain λ = 1e-4
    Shared 100 Hz command layer uses p_cmd = p_policy − λ f_ext with λ=10^{-4}; affects contact tracking for all methods.
  • MCD / SRL loss weights λ_MCD, λ_SRL
    Stage-3 distillation objective weights manifold consistency vs demonstration regression; values not numerically reported but required for the one-step high-frequency sampler.
  • indicator loss weights γ, α_MM and sample count N=8 = N=8; γ, α_MM unspecified
    Stage-2 confidence-weighted regression fits σ(o) from N low-frequency residual samples; controls when the gate fires.
  • frequency pair and horizons (10 Hz/16 vs 30 Hz/48, 1.6 s, slow refresh 1 s) = 10 Hz H=16; 30 Hz H=48; h_e=10/30
    Hand-chosen multi-frequency schedule defines what ‘adaptive’ means operationally and matches camera/force rates.
  • distillation timestep grid G = {99,79,59,39,19,0}
    Six-timestep grid {99,79,59,39,19,0} chosen for progressive one-step high-frequency distillation.
axioms (5)
  • domain assumption DDIM with η=0 yields a deterministic denoising trajectory for fixed initial noise, enabling cached slow context and consistent closed-loop force updates within a chunk.
    Imported from prior end-to-end visual-force diffusion (Preliminary; ImplicitRDP) and used as the low-frequency inference mechanism.
  • domain assumption Pre-contact phases are dominated by action multimodality measurable from visual tokens, while post-contact phases are dominated by force-constrained unimodal control needing higher update rate.
    Stated in Introduction/Fig. 1 and motivates indicator-gated switching; empirical support is correlational (Fig. 9).
  • ad hoc to paper Predicting clean actions on the robot action manifold with DDPM residual conversion is stabler to distill for one-step robot control than epsilon/score/velocity targets.
    Core MCD design claim (Sec. IV-C), inspired by JiT/Pixel Mean Flows but not independently proven outside these tasks.
  • domain assumption Causal action-force masking prevents future contact leakage while allowing parallel diffusion training and closed-loop force conditioning.
    Carried from the consistent reactive diffusion setup in Sec. III.
  • standard math Standard DDPM forward process and epsilon MSE supervision are valid training objectives when the network is reparameterized to predict actions.
    Eqs. 1–4; conventional diffusion training identity.
invented entities (3)
  • Multimodality indicator head σ(o) no independent evidence
    purpose: Estimate local action ambiguity from slow visual tokens and gate low- vs high-frequency samplers.
    Trained in Stage 2 from empirical low-frequency residuals; no external validated multimodality meter beyond this paper’s calibration set and task curves.
  • Manifold Consistency Distillation (MCD) no independent evidence
    purpose: Distill the high-frequency sampler into one-step action prediction on the robot action manifold with SRL regularization.
    New distillation objective in Sec. IV-C; evidence is the distillation ablation (Table III) on the same three tasks.
  • Frequency-adaptive positional encoding on a shared temporal grid no independent evidence
    purpose: Let one Transformer predict both sparse 10 Hz and dense 30 Hz action chunks without separate backbones.
    Architectural device in Sec. IV-A/Fig. 2; validated only via FA-RDP system results.

pith-pipeline@v1.2.0-daily-grok45 · 16951 in / 4010 out tokens · 68594 ms · 2026-07-31T02:43:18.460870+00:00 · methodology

0 comments
read the original abstract

In contact-rich manipulation, action multimodality and reactivity dominate different stages of a single episode. Before contact, multiple trajectories might be equally valid, making it important to preserve diverse action modes. After contact, geometric constraints and force limits narrow the solution space, while successful execution demands rapid responses to force feedback. However, standard diffusion policies use a fixed inference frequency and sampling steps throughout the episode, forcing a fundamental compromise: low-frequency, multi-step sampling better preserves pre-contact multimodality but responds slowly to force feedback, whereas high-frequency sampling improves reactivity but tends to collapse distinct pre-contact modes. To resolve this tradeoff, we present FA-RDP, a frequency-adaptive reactive diffusion policy. A shared multi-frequency visual-force Transformer predicts action chunks at both low and high frequencies, while a learned multimodality indicator dynamically selects multi-step low-frequency sampling before contact and one-step high-frequency sampling as action ambiguity decreases. We further introduce Manifold Consistency Distillation (MCD), which reparameterizes the diffusion network to predict actions on the robot action manifold while retaining DDPM-based residual supervision. Experiments on three contact-rich manipulation tasks show that FA-RDP achieves the highest success rate while preserving diverse pre-contact trajectory modes. Code and videos are available at https://fa-rdp.github.io.

Figures

Figures reproduced from arXiv: 2607.28596 by Cewu Lu, Chuan Wen, Han Xue, Jun Lv, Lifeng Zhuo, Shirun Tang, Wendi Chen.

Figure 1
Figure 1. Figure 1: Multimodality-guided frequency adaptation across manipulation [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Multi-frequency visual-force Transformer. The frequency mode [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Manifold consistency distillation for the high-frequency sampler. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Robot workspace and task objects. The Rizon 4s follower arm [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 9
Figure 9. Figure 9: Multimodality indicator value and force curves on the three contact [PITH_FULL_IMAGE:figures/full_fig_p006_9.png] view at source ↗
Figure 8
Figure 8. Figure 8: Pre-contact mode distributions across the three tasks. The high [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗
Figure 5
Figure 5. Figure 5: Dual Box Flipping failure-case comparison. DP and ImplicitRDP can lose contact during the flip, RDP can contact the wrong box region, and [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Dual Switch Toggling failure-case comparison. DP and ImplicitRDP can lose contact during the toggle, RDP can contact the wrong switch region, [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Dual Button Pressing failure-case comparison. DP and ImplicitRDP can lose contact during the press, RDP can contact the wrong button region, [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

34 extracted references · 18 linked inside Pith

  1. [1]

    A survey on imitation learning for contact-rich tasks in robotics,

    T. Tsuji, Y . Kato, G. Solak, H. Zhang, T. Petri ˇc, F. Nori, and A. Ajoudani, “A survey on imitation learning for contact-rich tasks in robotics,”The International Journal of Robotics Research, p. 02783649261417694, 2025

    2025

  2. [2]

    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

  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,”arXiv preprint arXiv:2503.02881, 2025

    Pith/arXiv arXiv 2025

  4. [4]

    Implicitrdp: An end-to-end visual-force diffusion policy with structural slow-fast learning,

    W. Chen, H. Xue, Y . Wang, F. Zhou, J. Lv, Y . Jin, S. Tang, C. Wen, and C. Lu, “Implicitrdp: An end-to-end visual-force diffusion policy with structural slow-fast learning,”IEEE Robotics and Automation Letters, 2026

    2026

  5. [5]

    Progressive distillation for fast sampling of diffusion models,

    T. Salimans and J. Ho, “Progressive distillation for fast sampling of diffusion models,”arXiv preprint arXiv:2202.00512, 2022

    Pith/arXiv arXiv 2022

  6. [6]

    Consistency policy: Accelerated visuomotor policies via consistency distillation,

    A. Prasad, K. Lin, J. Wu, L. Zhou, and J. Bohg, “Consistency policy: Accelerated visuomotor policies via consistency distillation,”arXiv preprint arXiv:2405.07503, 2024

    Pith/arXiv arXiv 2024

  7. [7]

    Learning fine-grained bimanual manipulation with low-cost hardware,

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

    Pith/arXiv arXiv 2023

  8. [8]

    3d diffusion policy: Generalizable visuomotor policy learning via simple 3d representations,

    Y . Ze, G. Zhang, K. Zhang, C. Hu, M. Wang, and H. Xu, “3d diffusion policy: Generalizable visuomotor policy learning via simple 3d representations,”arXiv preprint arXiv:2403.03954, 2024

    Pith/arXiv arXiv 2024

  9. [9]

    Dexforce: Extracting force-informed actions from kinesthetic demonstrations for dexterous manipulation,

    C. Chen, Z. Yu, H. Choi, M. Cutkosky, and J. Bohg, “Dexforce: Extracting force-informed actions from kinesthetic demonstrations for dexterous manipulation,”IEEE Robotics and Automation Letters, 2025

    2025

  10. [10]

    3d-vitac: Learning fine-grained manipulation with visuo-tactile sensing,

    B. Huang, Y . Wang, X. Yang, Y . Luo, and Y . Li, “3d-vitac: Learning fine-grained manipulation with visuo-tactile sensing,”arXiv preprint arXiv:2410.24091, 2024

    Pith/arXiv arXiv 2024

  11. [11]

    Factr: Force-attending curriculum training for contact-rich policy learning,

    J. J. Liu, Y . Li, K. Shaw, T. Tao, R. Salakhutdinov, and D. Pathak, “Factr: Force-attending curriculum training for contact-rich policy learning,”arXiv preprint arXiv:2502.17432, 2025

    Pith/arXiv arXiv 2025

  12. [12]

    Factr 2: Learning external force sensing for commodity robot arms improves policy learning,

    S. Oh, J. J. Liu, T. Tao, P. Han, K. Shaw, S. Funabashi, R. Salakhut- dinov, and D. Pathak, “Factr 2: Learning external force sensing for commodity robot arms improves policy learning,”arXiv preprint arXiv:2606.12406, 2026

    Pith/arXiv arXiv 2026

  13. [13]

    Forcevla: Enhancing vla models with a force-aware moe for contact-rich manipulation,

    J. Yu, H. Liu, Q. Yu, J. Ren, C. Hao, H. Ding, G. Huang, G. Huang, Y . Song, P. Cai,et al., “Forcevla: Enhancing vla models with a force-aware moe for contact-rich manipulation,”Advances in Neural Information Processing Systems, vol. 38, pp. 93 409–93 439, 2026

    2026

  14. [14]

    Dexterity from touch: Self-supervised pre-training of tactile representations with robotic play,

    I. Guzey, B. Evans, S. Chintala, and L. Pinto, “Dexterity from touch: Self-supervised pre-training of tactile representations with robotic play,”arXiv preprint arXiv:2303.12076, 2023

    Pith/arXiv arXiv 2023

  15. [15]

    Vtam: Video-tactile-action models for complex physical interaction beyond vlas,

    H. Yuan, W. Yi, Z. Zhang, W. Chen, Y . Mo, J. Yin, X. Li, X. Zeng, C. Wen, C. Lu,et al., “Vtam: Video-tactile-action models for complex physical interaction beyond vlas,”arXiv preprint arXiv:2603.23481, 2026

    arXiv 2026

  16. [16]

    Ftp-1: A generalist foundation tactile policy across tactile sensors for contact-rich manipulation,

    C. Yuan, Z. Zhang, M. Zhou, W. Chen, Y . Wang, Z. Liu, D. Niu, S. Wang, H. Zhang, W. Zhang,et al., “Ftp-1: A generalist foundation tactile policy across tactile sensors for contact-rich manipulation,” arXiv preprint arXiv:2606.13102, 2026

    Pith/arXiv arXiv 2026

  17. [17]

    Compliant residual dagger: Improving real-world contact-rich manipulation with human correc- tions,

    X. Xu, Y . Hou, Z. Liu, and S. Song, “Compliant residual dagger: Improving real-world contact-rich manipulation with human correc- tions,”Advances in Neural Information Processing Systems, vol. 38, pp. 139 559–139 581, 2026

    2026

  18. [18]

    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. He,et al., “Force policy: Learning hybrid force-position control policy under interaction frame for contact-rich manipulation,” arXiv preprint arXiv:2602.22088, 2026

    Pith/arXiv arXiv 2026

  19. [19]

    Hipolicy: Hierarchical multi-frequency action chunking for policy learning,

    J. Zhang, Z. Han, J. Wang, X. Wu, S. Lin, J. Li, H. Fan, R. Wu, D. Li, and H. Dong, “Hipolicy: Hierarchical multi-frequency action chunking for policy learning,”arXiv preprint arXiv:2604.06067, 2026

    Pith/arXiv arXiv 2026

  20. [20]

    Learning high- frequency continuous action chunks in latent space,

    K. Wang, Y . Zheng, Y . Zheng, J. Zhao, and W. Ding, “Learning high- frequency continuous action chunks in latent space,”arXiv preprint arXiv:2605.24931, 2026

    Pith/arXiv arXiv 2026

  21. [21]

    Denoising tells when to replan: Denoising-variance adaptive chunking for flow-based robot policies,

    X. Feng, Y . Cheng, C. Shi, B. Han, Y . Yan, Y . Hong, Z. Tian, and L. Jiang, “Denoising tells when to replan: Denoising-variance adaptive chunking for flow-based robot policies,”arXiv preprint arXiv:2606.03847, 2026

    Pith/arXiv arXiv 2026

  22. [22]

    Aha-wam: Asynchronous horizon-adaptive world-action modeling with observation-guided context routing,

    J. Cai, L. Ling, S. Chu, Z. Liu, J. Kang, Z. Liang, W. Xu, Y . Mao, W. Zhang, X. Yang,et al., “Aha-wam: Asynchronous horizon-adaptive world-action modeling with observation-guided context routing,”arXiv preprint arXiv:2606.09811, 2026

    Pith/arXiv arXiv 2026

  23. [23]

    Manip- force: Force-guided policy learning with frequency-aware representa- tion for contact-rich manipulation,

    G. Lee, Y . Lee, K. Kim, S. Lee, S. Noh, S. Back, and K. Lee, “Manip- force: Force-guided policy learning with frequency-aware representa- tion for contact-rich manipulation,”arXiv preprint arXiv:2509.19047, 2025

    arXiv 2025

  24. [24]

    Denoising diffusion implicit models,

    J. Song, C. Meng, and S. Ermon, “Denoising diffusion implicit models,”arXiv preprint arXiv:2010.02502, 2020

    Pith/arXiv arXiv 2010

  25. [25]

    Consistency models,

    Y . Song, P. Dhariwal, M. Chen, and I. Sutskever, “Consistency models,” 2023

    2023

  26. [26]

    Manicm: Real-time 3d diffusion policy via consistency model for robotic manipulation,

    G. Lu, Z. Gao, T. Chen, W. Dai, Z. Wang, W. Ding, and Y . Tang, “Manicm: Real-time 3d diffusion policy via consistency model for robotic manipulation,”arXiv preprint arXiv:2406.01586, 2024

    Pith/arXiv arXiv 2024

  27. [27]

    Hybrid consistency policy: Decoupling multi-modal diversity and real-time efficiency in robotic manipulation,

    Q. Zhao, Y . Shen, X. Zhai, D. Wu, J. Qi, C. Hao, J. Hu, and Q. Yu, “Hybrid consistency policy: Decoupling multi-modal diversity and real-time efficiency in robotic manipulation,”IEEE Robotics and Automation Letters, 2026

    2026

  28. [28]

    Mean flows for one- step generative modeling,

    Z. Geng, M. Deng, X. Bai, Z. Kolter, and K. He, “Mean flows for one- step generative modeling,”Advances in Neural Information Processing Systems, vol. 38, pp. 75 460–75 482, 2026

    2026

  29. [29]

    Mp1: Meanflow tames policy learning in 1-step for robotic manipulation,

    J. Sheng, Z. Wang, P. Li, and M. Liu, “Mp1: Meanflow tames policy learning in 1-step for robotic manipulation,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 40, no. 22, 2026, pp. 18 532–18 539

    2026

  30. [30]

    Freqpolicy: Efficient flow-based visuomotor policy via frequency consistency,

    Y . Su, N. Liu, D. Chen, Z. Zhao, K. Wu, M. Li, Z. Xu, Z. Che, and J. Tang, “Freqpolicy: Efficient flow-based visuomotor policy via frequency consistency,”Advances in Neural Information Processing Systems, vol. 38, pp. 27 769–27 797, 2026

    2026

  31. [31]

    Back to basics: Let denoising generative models denoise,

    T. Li and K. He, “Back to basics: Let denoising generative models denoise,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026, pp. 36 115–36 125

    2026

  32. [32]

    One-step latent-free image generation with pixel mean flows,

    Y . Lu, S. Lu, Q. Sun, H. Zhao, Z. Jiang, X. Wang, T. Li, Z. Geng, and K. He, “One-step latent-free image generation with pixel mean flows,”arXiv preprint arXiv:2601.22158, 2026

    Pith/arXiv arXiv 2026

  33. [33]

    Vggt: Visual geometry grounded transformer,

    J. Wang, M. Chen, N. Karaev, A. Vedaldi, C. Rupprecht, and D. Novotny, “Vggt: Visual geometry grounded transformer,” inPro- ceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 5294–5306

    2025

  34. [34]

    Rethinking camera choice: An empirical study on fisheye camera properties in robotic manipulation,

    H. Xue, N. Min, X. Liu, W. Chen, Y . Fang, J. Lv, C. Lu, and C. Wen, “Rethinking camera choice: An empirical study on fisheye camera properties in robotic manipulation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026, pp. 35 059–35 069

    2026