Pith. sign in

REVIEW 4 major objections 4 minor 31 references

This paper claims that initializing a generative flow at the robot's recent state history, instead of Gaussian noise, makes action generation faster and cheaper while preserving success rates.

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

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

T0 review · deepseek-v4-flash

2026-08-03 06:10 UTC pith:MMCQLOW7

load-bearing objection A promising but not yet proven application of history-initialized stochastic interpolants to robotic LfD: correct math, competitive results, but the headline latency advantage is confounded by a 15x smaller network. the 4 major comments →

arxiv 2607.29482 v1 pith:MMCQLOW7 submitted 2026-07-31 cs.RO

Temporal Policy: History-Initialized Action Generation for Robotic Learning from Demonstration

Dylan Miller , Martin Jagersand This is my paper
classification cs.RO
keywords robotic learning from demonstrationstochastic interpolantsflow matchingdata-dependent couplingaction chunkingtransport costinference latencygenerative policy
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.

Temporal Policy reframes robot action generation as a transport problem: instead of mapping Gaussian noise to future actions, the generative flow starts from the robot's recent state history and carries it to the future action chunk. Because past and future states in smooth demonstrations are correlated, the expected transport distance shrinks and the learned vector field becomes nearly straight, so the policy needs very few sampling steps. The paper reports 19.1 ms inference latency on a consumer GPU with ten function evaluations, overall matching the success rates of noise-initialized diffusion and flow-matching baselines across simulation benchmarks and a physical 7-DoF teleoperation task. If this holds, generative policies become practical for high-frequency closed-loop control without specialized hardware. The same trained model also supports both deterministic and stochastic sampling at inference.

Core claim

On its own terms, the paper's discovery is that a data-dependent coupling — using the robot's recent state history as the source and the future action chunk as the target — turns action generation into a short, nearly straight point-to-distribution transport instead of a long curved journey from Gaussian noise. With smooth demonstrations the history and the near-future actions are positively correlated, so the expected squared distance the flow must cover is strictly smaller than in the independent-Gaussian setup. Training a stochastic-interpolant drift with this coupling produces straight vector fields, which is why the policy can generate good actions in one to ten solver steps. On a preci

What carries the argument

The central object is the history-initialized stochastic interpolant: a continuous probability path x_lambda = (1-lambda)x0 + lambda x1 + epsilon(1-lambda)w_lambda that mixes source history, target action chunk, and Wiener noise, plus a drift network trained by square-loss regression to approximate the target velocity. The source choice does the load-bearing work: because the expected cross-correlation E[x1^T x0] is positive for smooth demonstrations, the expected squared transport cost J = 2HD - 2E[x1^T x0] is strictly smaller than the Gaussian baseline's 2HD. Straightness follows from the linear interpolant combined with the short source-target distance, and analytic score recovery lets th

Load-bearing premise

The entire efficiency gain rests on the assumption that a robot's recent state history and its immediate future action sequence are strongly positively correlated, i.e., that demonstrations are smooth, purposeful motions; with pauses, reversals, or high-frequency jitter the cross-correlation E[x1^T x0] shrinks and the transport-cost and straightness advantages erode.

What would settle it

Take any demonstration dataset, randomly shuffle the pairing of history windows and future action chunks to break temporal correlation while keeping marginals, retrain the same Temporal Policy, and measure transport cost and the number of function evaluations needed for a given success rate. If the low-NFE advantage persists despite zero or negative cross-correlation, then the temporal coupling is not the cause. Alternatively, compute E[x1^T x0] directly on a jittery dataset and show the predicted cost reduction J_temporal < J_indep no longer holds.

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

If this is right

  • Generative visuomotor policies can operate at roughly 50 Hz on a single consumer GPU (19.1 ms inference), making them viable for closed-loop control rather than open-loop re-planning.
  • A single trained model supports both fast deterministic (ODE) sampling for latency-critical control and stochastic (SDE) sampling for diversity, with the noise schedule adjustable at inference time.
  • The approach is architecture-agnostic, so it can combine with other generative acceleration techniques such as consistency distillation or higher-order ODE solvers.
  • Fewer demonstrations are needed to reach a given success rate, because the source-target gap is smaller and the learning problem is simpler.
  • History initialization alone yields usable actions at a single function evaluation, suggesting the learned vector field is nearly straight and well behaved.

Where Pith is reading between the lines

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

  • Editorial inference: Because the dynamic transport-cost reduction is demonstrated on only one task, the broad claim that low-NFE success follows from the coupling is plausible but not yet fully tested; measuring transport cost on the multi-human, high-variance datasets would directly probe the correlation assumption.
  • Editorial inference: The requirement that states and actions share a representation limits the method to tasks where actions are positions or poses; extending to force/torque or velocity commands would need a different source construction, and the paper's own real-world experiment used gripper velocities, a representation mismatch.
  • Editorial inference: If the benefit is mostly 'starting closer,' then initializing from the previous action chunk or a velocity-projected state, as the authors suggest for future work, may yield similar gains; a head-to-head comparison would isolate the source-coupling effect from the interpolant's straightness.
  • Editorial inference: The authors themselves question whether history initialization preserves full multimodality rather than collapsing it; a behavioral-diversity metric on multi-human datasets would settle whether the method retains the range of valid demonstration strategies.

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

4 major / 4 minor

Summary. Temporal Policy proposes a generative policy for robot learning from demonstration based on stochastic interpolants. Instead of initializing the generative flow with independent Gaussian noise, the method initializes it with the robot's recent state history, coupling past states to future action chunks. The paper derives the interpolant, drift objective, and score-recovery equations, presents training and inference algorithms, and evaluates the method on five Robomimic tasks and a real Barrett WAM teleoperation platform. It reports competitive success rates, reduced transport cost, low NFE, and a 19.1 ms inference latency on an RTX 4080, alongside data-efficiency and real-world results.

Significance. If the central claims hold, the paper addresses a real bottleneck in deploying generative policies for closed-loop control: inference latency. The data-dependent coupling idea is well motivated, and the SDE/ODE derivation in Section III-C is correct. The paper also provides public code and reports success-rate parity with strong baselines across multiple tasks, which is valuable. However, the headline efficiency claim is not architecture-controlled, and the default deterministic inference mode has a fundamental limitation for a point-mass source that is not discussed in the proper framework. The transport-cost claim is supported by only one task and relies on an unmeasured correlation assumption. These issues are load-bearing for the paper's main contributions, so the manuscript requires substantive revision rather than minor cleanup.

major comments (4)
  1. [IV-A, Table I] The efficiency comparison is not architecture-controlled. Temporal Policy uses 17M parameters while Diffusion Policy and CFM use 255M, so the 19.1 ms latency improvement could be largely due to model size. The paper states that a scaled-down 17M CFM baseline was trained on Transport and did not degrade, but it does not report that baseline's success rates, NFE-success curve, or inference latency. To support the claim that history initialization enables low NFE and fast inference, the authors should provide a same-backbone, same-size noise-initialized CFM baseline evaluated across tasks at NFE 1-10, with latency numbers.
  2. [III-E, Algorithm 2, Eq. (6)] The recommended deterministic sampling mode (g=0) cannot generate a distribution over actions from a fixed history. A deterministic ODE maps a point to a point; starting from x0 yields a single output trajectory, not a sample from a multimodal action distribution. This is especially important because the paper motivates generative models by their ability to capture multimodal LfD behaviors. The authors should either evaluate and report the stochastic SDE mode at matched NFE, or explicitly reframe the default policy as a deterministic conditional-mean-style policy and provide evidence that this preserves the useful multimodality (e.g., via success metrics or diversity measures). The current text does not resolve this issue.
  3. [III-B, IV-B] The transport-cost reduction is claimed as a general property, but the static argument assumes E[x1^T x0] > 0 and the dynamic transport cost is measured only on Square ph (Table II). Expert demonstrations with pauses, reversals, or jitter can make this correlation small or negative, so the geometric benefit is task-dependent. The authors should report the empirical cross-correlation between source and target for each task, and ideally dynamic transport costs across all five tasks. Without this, the 'nearly an order of magnitude' reduction is not established beyond a single task.
  4. [IV-E] The real-world experiment appears to violate the structural equivalence S ~= A required in Section III-A. The state-action coupling is described as using follower joint positions and gripper velocities, but the formulation assumes both source and target live in the same configuration space (joint positions or Cartesian poses). Please clarify the exact representation of x0 and x1 in the real-world experiment, and explain how gripper velocities fit the S ~= A requirement, or relax the formalism to accommodate them. This matters because the real-world deployment is used as evidence that the method works outside simulation.
minor comments (4)
  1. [IV-C, Fig. 3] The symbol ε is used both for the training noise scale in Eq. (1) and for the inference noise schedule in Eq. (6). In the ablation, ε is apparently used as gλ without a clear definition. Please distinguish these two quantities.
  2. [Table I] The table reports Max/Average success rates but no variance or confidence intervals, despite stating that values are averaged across 3 seeds. Reporting standard errors would help assess the significance of the small differences between methods.
  3. [Fig. 4] The data-efficiency plot appears to be for Square ph only and no error bars are shown. Clarify this in the caption and text, and consider adding error bars or additional tasks.
  4. [Eq. (5)] The score formula divides by ε^2 λ(1-λ), so the expression is singular at λ=0 and λ=1 when ε=0. The text should state how the deterministic ε=0 case is handled numerically, especially for the first Euler step.

Circularity Check

0 steps flagged

No significant circularity: transport-cost reduction is a stated design property and empirically measured, not a fitted prediction; self-citations are non-load-bearing.

full rationale

The paper's derivation is self-contained relative to the circularity test. The central analytical step is the static transport-cost identity in Sec III-B, J = 2HD - 2E[x1^T x0]. This is an algebraic identity, and the claimed inequality J_temporal < J_indep follows from the explicitly stated assumption that expert demonstrations are smooth enough to give E[x1^T x0] > 0. No parameter is fitted to data and then renamed as a prediction; the transport cost and straightness are subsequently measured (Table II). The 'straight vector fields' component is partly a design choice through the linear interpolant of Eq. 1, and the paper transparently states in Sec IV-B that its lower transport cost 'mostly stems from starting closer to the target manifold, rather than from straightening the flow.' The method relies on external stochastic-interpolant theory [6,7] from non-overlapping authors; the authors' own prior work [2] appears only as a general example of video-based LfD and is not load-bearing. There is no invoked uniqueness theorem from the authors themselves. The 19.1 ms latency claim is confounded by the 15x parameter-count difference versus the 255M-parameter baselines, and the scaled-down CFM check is mentioned without reported numbers; these are experimental-design and evidence concerns, not circularity. No load-bearing step reduces by construction or self-citation to its own input.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The paper introduces no new physical entities or forces. Its theoretical load is carried by prior stochastic-interpolant results and by an empirically plausible but unmeasured assumption about temporal correlation in demonstrations. The free parameters are standard hyperparameters of the generative setup, with the model size being a notable confound in the efficiency claim.

free parameters (4)
  • noise scale epsilon = not reported; ablation tested 0.0, 0.25, 0.5, 0.75, 1.0
    Controls the stochastic noise in the interpolant (Eq. 1) and the inference SDE; chosen by hand, not optimized. The main results use deterministic ODE at inference.
  • horizon H = not reported for simulation; 16 in real-world (execute 8)
    Length of the action chunk; affects the transport distance and the temporal coupling; a design choice not sweeped in the paper.
  • source-target overlap d = 2 in real-world; not reported in simulation
    Shift between the state history source and the future action target; affects correlation and discontinuities; chosen by hand.
  • model size (17M parameters) = 17M
    Chosen because the full 255M model overfit for Temporal Policy; the latency comparison is confounded because baselines use 255M.
axioms (4)
  • domain assumption Stochastic interpolant framework of Albergo et al. [6] is valid and applicable to arbitrary source and target distributions.
    The entire method is built on this framework; the paper does not prove it.
  • domain assumption Analytic score recovery of Chen et al. [7] holds for the chosen interpolant and allows noise-schedule modification at inference.
    Used to derive Eq. 5 and Eq. 6; accepted from prior work.
  • domain assumption Robot state space and action space are isomorphic (S ≈ A ≈ R^D) and the action chunk has the same representation as the state history.
    Section III-A imposes this equality; required for the temporal coupling to be a transport between like spaces.
  • domain assumption Expert demonstrations have strong positive cross-correlation between state history and near-future actions (E[x1^T x0] > 0).
    Section III-B uses this to claim J_temporal < J_indep; not measured and may fail for pauses, reversals, or high-frequency jitter.

pith-pipeline@v1.3.0-daily-deepseek · 11371 in / 15021 out tokens · 170412 ms · 2026-08-03T06:10:46.250648+00:00 · methodology

0 comments
read the original abstract

By relying on independent couplings from uninformative Gaussian priors, standard diffusion and flow matching models are forced to learn complex, high-cost vector fields to reach the physical action space. Generative models excel at capturing multimodal behaviors for robotic Learning from Demonstration (LfD), but often suffer from high inference cost. This paper introduces Temporal Policy, a generative framework based on stochastic interpolants that formulates action generation as a temporally coupled transport problem. By initializing the generative flow at the robot's recent history, we explicitly couple past states to future action sequences. This data-dependent coupling reduces transport cost and produces straight vector fields. We validate Temporal Policy across visuomotor simulation benchmarks and on a physical Barrett WAM 2x 7DoF teleoperation platform. Our approach reduces transport costs by nearly an order of magnitude compared to noise-initialized baselines, achieving a 19.1 ms inference latency on a single NVIDIA RTX 4080. Crucially, these geometric and computational efficiencies are achieved while matching the success rates of state-of-the-art baselines. This simplified transport geometry bypasses the computational bottleneck of independent Gaussian priors, helping enable high-frequency, closed-loop control. The code is publicly available at https://github.com/dmiller12/TemporalPolicy.

Figures

Figures reproduced from arXiv: 2607.29482 by Dylan Miller, Martin Jagersand.

Figure 1
Figure 1. Figure 1: Conceptual comparison of generative modeling approaches. (1) [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of our architecture. Unlike standard generative models that initialize from independent Gaussian noise [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of model performance on the Tool Hang [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: illustrates success rates as a function of dataset size. Temporal Policy consistently outperforms Diffusion Policy across all data regimes. We hypothesize that the Temporal Policy formulation simplifies the learning problem by reducing the discrepancy between the source and target distributions, allowing the model to generalize effectively with fewer demonstrations. This efficiency has important practical … view at source ↗
Figure 5
Figure 5. Figure 5: Overview of the mug-hang task. (a) The initial [PITH_FULL_IMAGE:figures/full_fig_p007_5.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

31 extracted references · 3 linked inside Pith

  1. [1]

    A survey of robot learning from demonstration,

    B. D. Argall, S. Chernova, M. Veloso, and B. Browning, “A survey of robot learning from demonstration,”Robotics and Autonomous Systems, vol. 57, no. 5, pp. 469–483, May 2009. [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/S0921889008001772

    2009

  2. [2]

    Generalizable task representation learning from human demonstration videos: A geometric approach,

    J. Jin and M. Jagersand, “Generalizable task representation learning from human demonstration videos: A geometric approach,” in 2022 International Conference on Robotics and Automation (ICRA). Philadelphia, PA, USA: IEEE, May 2022, pp. 2504–2510. [Online]. Available: https://ieeexplore.ieee.org/document/9812195/

    arXiv 2022

  3. [3]

    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, Sept. 2025. [Online]. Available: https://journals.sagepub.com/doi/10.1177/02783649241273668

    doi:10.1177/02783649241273668 2025

  4. [4]

    Flow Matching for Generative Modeling,

    Y . Lipman, R. T. Q. Chen, H. Ben-Hamu, M. Nickel, and M. Le, “Flow Matching for Generative Modeling,” inThe Eleventh International Conference on Learning Representations, 2023. [Online]. Available: https://openreview.net/forum?id=PqvMRDCJT9t

    2023

  5. [5]

    Denoising Diffusion Probabilistic Models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising Diffusion Probabilistic Models,” inAdvances in Neural Information Processing Systems, vol. 33. Curran Associates, Inc., 2020, pp. 6840–6851. [Online]. Available: https://proceedings.neurips.cc/paper/2020/hash/ 4c5bcfec8584af0d967f1ab10179ca4b-Abstract.html

    2020

  6. [6]

    Stochastic Interpolants: A Unifying Framework for Flows and Diffusions,

    M. Albergo, N. M. Boffi, and E. Vanden-Eijnden, “Stochastic Interpolants: A Unifying Framework for Flows and Diffusions,” Journal of Machine Learning Research, vol. 26, no. 209, pp. 1–80,

  7. [7]

    Probabilistic Forecasting with Stochastic Interpolants and F ¨ollmer Processes,

    Y . Chen, M. Goldstein, M. Hua, M. S. Albergo, N. M. Boffi, and E. Vanden-Eijnden, “Probabilistic Forecasting with Stochastic Interpolants and F ¨ollmer Processes,” inProceedings of the 41st International Conference on Machine Learning. PMLR, July 2024, pp. 6728–6756. [Online]. Available: https: //proceedings.mlr.press/v235/chen24n.html

    2024

  8. [8]

    Goal-Conditioned Imitation Learning using Score-based Diffusion Policies,

    M. Reuss, M. Li, X. Jia, and R. Lioutikov, “Goal-Conditioned Imitation Learning using Score-based Diffusion Policies,” inRobotics: Science and Systems, 2023. [Online]. Available: http://arxiv.org/abs/ 2304.02532

    Pith/arXiv arXiv 2023

  9. [9]

    Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow,

    X. Liu, C. Gong, and Q. Liu, “Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow,” inThe Eleventh International Conference on Learning Representations, Sept. 2022. [Online]. Available: https://openreview.net/forum?id=XVjTT1nw5z

    2022

  10. [10]

    Consistency models,

    Y . Song, P. Dhariwal, M. Chen, and I. Sutskever, “Consistency models,” inProceedings of the 40th International Conference on Machine Learning, ser. ICML’23, vol. 202. Honolulu, Hawaii, USA: JMLR.org, July 2023, pp. 32 211–32 252

    2023

  11. [11]

    ActionFlow: Equivariant, Accurate, and Efficient Policies with Spatially Symmetric Flow Matching,

    N. Funk, J. Urain, J. Carvalho, V . Prasad, G. Chalvatzaki, and J. Peters, “ActionFlow: Equivariant, Accurate, and Efficient Policies with Spatially Symmetric Flow Matching,” Sept. 2024. [Online]. Available: http://arxiv.org/abs/2409.04576

    Pith/arXiv arXiv 2024

  12. [12]

    $π 0$: A Vision-Language-Action Flow Model for General Robot Control,

    K. Black, N. Brown, D. Driess, A. Esmail, M. Equi, C. Finn, N. Fusai, L. Groom, K. Hausman, B. Ichter, S. Jakubczak, T. Jones, L. Ke, S. Levine, A. Li-Bell, M. Mothukuri, S. Nair, K. Pertsch, L. X. Shi, J. Tanner, Q. Vuong, A. Walling, H. Wang, and U. Zhilinsky, “$π 0$: A Vision-Language-Action Flow Model for General Robot Control,” inRobotics: Science an...

    2025

  13. [13]

    Flow Policy: Generalizable Visuomotor Policy Learning via Flow Matching,

    Y . Fang, X. Zhang, H. Cheng, X. Zang, R. Song, and J. Zhao, “Flow Policy: Generalizable Visuomotor Policy Learning via Flow Matching,”IEEE/ASME Transactions on Mechatronics, vol. 31, no. 1, pp. 140–150, Feb. 2026. [Online]. Available: https: //ieeexplore.ieee.org/abstract/document/11090155

    arXiv 2026

  14. [14]

    ManiFlow: A General Robot Manipulation Policy via Consistency Flow Training,

    G. Yan, J. Zhu, Y . Deng, S. Yang, R.-Z. Qiu, X. Cheng, M. Memmel, R. Krishna, A. Goyal, X. Wang, and D. Fox, “ManiFlow: A General Robot Manipulation Policy via Consistency Flow Training,” inProceedings of The 9th Conference on Robot Learning. PMLR, Oct. 2025, pp. 2268–2293. [Online]. Available: https://proceedings.mlr.press/v305/yan25a.html

    2025

  15. [15]

    Multisample Flow Matching: Straightening Flows with Minibatch Couplings,

    A.-A. Pooladian, H. Ben-Hamu, C. Domingo-Enrich, B. Amos, Y . Lipman, and R. T. Q. Chen, “Multisample Flow Matching: Straightening Flows with Minibatch Couplings,” inProceedings of the 40th International Conference on Machine Learning. PMLR, July 2023, pp. 28 100–28 127. [Online]. Available: https: //proceedings.mlr.press/v202/pooladian23a.html

    2023

  16. [16]

    Improving and generalizing flow-based generative models with minibatch optimal transport,

    A. Tong, K. Fatras, N. Malkin, G. Huguet, Y . Zhang, J. Rector-Brooks, G. Wolf, and Y . Bengio, “Improving and generalizing flow-based generative models with minibatch optimal transport,”Transactions on Machine Learning Research, p. 34, Mar. 2024. [Online]. Available: https://openreview.net/forum?id=CD9Snc73AW

    2024

  17. [17]

    Fast Flow-based Visuomotor Policies via Conditional Optimal Transport Couplings,

    A. Sochopoulos, N. Malkin, N. Tsagkas, J. Moura, M. Gienger, and S. Vijayakumar, “Fast Flow-based Visuomotor Policies via Conditional Optimal Transport Couplings,” inProceedings of The 9th Conference on Robot Learning. PMLR, Oct. 2025, pp. 3357–3377. [Online]. Available: https://proceedings.mlr.press/v305/sochopoulos25a.html

    2025

  18. [18]

    Diffusion Schr¨odinger Bridge with Applications to Score-Based Generative Modeling,

    V . De Bortoli, J. Thornton, J. Heng, and A. Doucet, “Diffusion Schr¨odinger Bridge with Applications to Score-Based Generative Modeling,” inAdvances in Neural Information Processing Systems, vol. 34. Curran Associates, Inc., 2021, pp. 17 695–17 709. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/2021/hash/ 940392f5f32a7ade1cc201767...

    2021

  19. [19]

    Likelihood Training of Schr¨odinger Bridge using Forward-Backward SDEs Theory,

    T. Chen, G.-H. Liu, and E. Theodorou, “Likelihood Training of Schr¨odinger Bridge using Forward-Backward SDEs Theory,” in International Conference on Learning Representations, Jan. 2022. [Online]. Available: https://openreview.net/forum?id=nioAdKCEdXB

    2022

  20. [20]

    Simulation-Free Schr ¨odinger Bridges via Score and Flow Matching,

    A. Y . Tong, N. Malkin, K. Fatras, L. Atanackovic, Y . Zhang, G. Huguet, G. Wolf, and Y . Bengio, “Simulation-Free Schr ¨odinger Bridges via Score and Flow Matching,” inProceedings of The 27th International Conference on Artificial Intelligence and Statistics. PMLR, Apr. 2024, pp. 1279–1287. [Online]. Available: https://proceedings.mlr.press/v238/tong24a.html

    2024

  21. [21]

    Stochastic Interpolants with Data-Dependent Couplings,

    M. S. Albergo, M. Goldstein, N. M. Boffi, R. Ranganath, and E. Vanden-Eijnden, “Stochastic Interpolants with Data-Dependent Couplings,” inProceedings of the 41st International Conference on Machine Learning. PMLR, July 2024, pp. 921–937. [Online]. Available: https://proceedings.mlr.press/v235/albergo24a.html

    2024

  22. [22]

    STFlow: Data-Coupled Flow Matching for Geometric Trajectory Simulation,

    K. B. ten Brinke, K. Minartz, and V . Menkovski, “STFlow: Data-Coupled Flow Matching for Geometric Trajectory Simulation,” May 2026. [Online]. Available: http://arxiv.org/abs/2505.18647

    Pith/arXiv arXiv 2026

  23. [23]

    Trajectory Flow Matching with Applications to Clinical Time Series Modelling,

    X. Zhang, Y . Pu, Y . Kawamura, A. Loza, Y . Bengio, D. L. Shung, and A. Tong, “Trajectory Flow Matching with Applications to Clinical Time Series Modelling,”Advances in Neural Information Processing Systems, vol. 37, pp. 107 198–107 224, Dec. 2024. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/2024/hash/ c1e67cde895c3c91edb43569ad...

    2024

  24. [24]

    Streaming Flow Policy: Simplifying diffusion/flow- matching policies by treating action trajectories as flow trajectories,

    S. Jiang, X. Fang, N. Roy, T. Lozano-P ´erez, L. P. Kaelbling, and S. Ancha, “Streaming Flow Policy: Simplifying diffusion/flow- matching policies by treating action trajectories as flow trajectories,” inProceedings of The 9th Conference on Robot Learning. PMLR, Oct. 2025, pp. 238–257. [Online]. Available: https: //proceedings.mlr.press/v305/jiang25a.html

    2025

  25. [25]

    VITA: Vision-to-Action Flow Matching Policy,

    D. Gao, B. Zhao, A. Lee, I. Chuang, H. Zhou, H. Wang, Z. Zhao, J. Zhang, and I. Soltani, “VITA: Vision-to-Action Flow Matching Policy,” Mar. 2026. [Online]. Available: http: //arxiv.org/abs/2507.13231

    arXiv 2026

  26. [26]

    U-Net: Convolutional Networks for Biomedical Image Segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-Net: Convolutional Networks for Biomedical Image Segmentation,” inMedical Image Computing and Computer-Assisted Intervention – MICCAI 2015, N. Navab, J. Hornegger, W. M. Wells, and A. F. Frangi, Eds. Cham: Springer International Publishing, 2015, pp. 234–241

    2015

  27. [27]

    Deep Residual Learning for Image Recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 770–778. [Online]. Available: https://openaccess.thecvf.com/content cvpr 2016/ html/He Deep Residual Learning CVPR 2016 paper.html

    2016

  28. [28]

    Deep spatial autoencoders for visuomotor learning,

    C. Finn, Xin Yu Tan, Yan Duan, T. Darrell, S. Levine, and P. Abbeel, “Deep spatial autoencoders for visuomotor learning,” in 2016 IEEE International Conference on Robotics and Automation (ICRA). Stockholm, Sweden: IEEE, May 2016, pp. 512–519. [Online]. Available: http://ieeexplore.ieee.org/document/7487173/

    arXiv 2016

  29. [29]

    FiLM: Visual Reasoning with a General Conditioning Layer,

    E. Perez, F. Strub, H. de Vries, V . Dumoulin, and A. Courville, “FiLM: Visual Reasoning with a General Conditioning Layer,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 32, no. 1, Apr. 2018. [Online]. Available: https://ojs.aaai.org/ index.php/AAAI/article/view/11671

    2018

  30. [30]

    What Matters in Learning from Offline Human Demonstrations for Robot Manipulation,

    A. Mandlekar, D. Xu, J. Wong, S. Nasiriany, C. Wang, R. Kulkarni, L. Fei-Fei, S. Savarese, Y . Zhu, and R. Mart ´ın-Mart´ın, “What Matters in Learning from Offline Human Demonstrations for Robot Manipulation,” inProceedings of the 5th Conference on Robot Learning. PMLR, Jan. 2022, pp. 1678–1690. [Online]. Available: https://proceedings.mlr.press/v164/mand...

    2022

  31. [2025]

    Available: http://jmlr.org/papers/v26/23-1605.html

    [Online]. Available: http://jmlr.org/papers/v26/23-1605.html