Pith. sign in

REVIEW 5 major objections 8 minor 42 references

Improving Trajectory Stitching with Flow Models

T0 review · 5 major / 8 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read This paper argues that generative trajectory planners fail on novel start-goal pairs because they select, not stitch, and presents Flow Planner, which uses a time-local UNet, action-noise augmentation, and trajectory splitting to compose…

desk verdict A genuinely useful empirical paper on making diffusion/flow planners stitch sub-trajectories, with a well-supported stitching result and an under-evidenced obstacle-avoidance headline. read the letter →

arxiv 2505.07802 v2 pith:DAUZLZ4C submitted 2025-05-12 cs.RO cs.AIcs.LG

classification cs.ROcs.AIcs.LG
keywords flowmatchingtrajectorystitchingrobotmanipulationguidedplanningobstacleavoidancediffusionmodelsforlocalreceptivefieldactionnoiseaugmentation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that generative trajectory planners fail on novel start-goal pairs because they select whole trajectories from the training set instead of stitching sub-trajectories. It proposes Flow Planner, built from three ingredients: a UNet with time-local (inpainting) conditioning, action-noise augmentation of the training data, and trajectory splitting during training and inference. On a Franka Panda arm, the method generates feasible plans for out-of-distribution boundary conditions and avoids spherical obstacles up to four times larger than the Motion Planning Diffusion baseline and other comparators, in simulation and on real hardware. The claim matters because it suggests the combinatorial space of motion clips needed for flexible behavior shrinks exponentially once a model can compose, rather than retrieve.

What carries the argument

The load-bearing mechanism is the local receptive field: each time step in the trajectory is predicted only from its temporal neighborhood, implemented by using a UNet with inpainting conditioning, as opposed to a transformer or direct conditioning that lets global start or goal information bias every state. Around this sit two auxiliary mechanisms. Action-noise augmentation adds Gaussian noise independently to each joint's torques during dataset collection, breaking state-to-state correlations so the model can switch between nearby overlapping clips. Trajectory splitting mixes half-length trajectories into training batches with 50 percent probability to prevent mode collapse, and at inference re-noises an initial plan up to 50 percent, splits it in half, and denoises each half separately with the midpoint as a new boundary condition to reduce dynamic inconsistency; trajectory lengths must remain powers of two because of the UNet's structure.

What would settle it

Move the inpainted goal far away while holding the local neighborhood of a given time step fixed, and measure how much that state's predicted value changes: if the output shifts substantially, the receptive field is not local, and the locality explanation is falsified; equivalently, inspect the trained UNet's effective receptive field with gradient-based sensitivity analysis.

Watch

Extended reading notes

Core claim

The central discovery is that stitching failure in generative trajectory planners is not inherent to flow matching but is driven by three identifiable deficiencies, each with a fix. A model whose conditioning is non-local in time, such as a transformer or direct conditioning, learns to reproduce whole dataset trajectories, so novel boundary conditions produce inconsistent plans; a UNet conditioned by inpainting, with a local receptive field, instead maximizes local consistency and can compose new trajectories. Dataset augmentation with action-specific Gaussian noise breaks correlations between joint states and enables the model to jump between overlapping clips, whereas randomizing start or goal positions and injecting random forces do not. Finally, mixing half-length trajectories into training batches prevents a guidance-induced mode collapse, and an inference-time split, re-noising to 50 percent and denoising each half separately with the midpoint as a new boundary condition, slows the growth of dynamic inconsistency under strong guidance. Together these techniques yield Flow Planner, which outperforms diffusion and VAE baselines in stitching error and obstacle-avoidance radius, and the result transfers to a real Franka Panda arm.

Load-bearing premise

The argument rests on the claim that the UNet's receptive field is completely local in time, so inpainting conditioning never leaks the global start or goal into a state's prediction; if downsampling gives the UNet a non-local receptive field, the stated mechanism behind the stitching improvement is not established.

Editorial extensions

If this is right

  • Planners built this way no longer need dense coverage of every start-goal pair; the paper states the combinatorial space of required clips reduces exponentially once stitching works.
  • Larger guidance scales become usable with inference-time splitting, so obstacle-avoidance and other cost-guided tasks can demand bigger deviations from the unguided plan without the trajectory becoming dynamically inconsistent.
  • The same architecture and conditioning choices transfer from simulation to real hardware, since the obstacle-avoidance task succeeds on a real Franka Panda with a box inside the maximum sphere.
  • Stitching performance is quantifiable: on a cross-shaped Franka dataset, Flow Planner reports a stitching error of 0.16 ± 0.07, versus 1.00 ± 0.97 for the Motion Planning Diffusion baseline.
  • Action-noise augmentation is the dataset-side key: same-noise, random-start/goal, and random-force schemes all leave stitching error higher than the 0.16 of Flow Planner.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the local-receptive-field explanation is right, then transformer-based planners should be stitchable by designing attention masks that are truly local; the paper's unsuccessful tridiagonal mask suggests this is a concrete open problem rather than a settled impossibility.
  • The paper's correlation-breaking hypothesis predicts a testable extension: augmentations that decorrelate successive joint states in a dimension-wise way, if they are less destructive than full torque noise, should preserve the smoothing-safety trade-off while enabling stitching.
  • Inference-time splitting is a generic guidance-stabilization trick, so it could be transplanted to other generator-matching planners, including diffusion planners, wherever strong guidance causes endpoint inconsistency.
  • Because the paper identifies architectural locality rather than flow matching itself as the cause, related stitching results in offline reinforcement learning and model-based planning might be re-examined through the same lens.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 8 minor

Summary. The paper proposes Flow Planner, a flow-matching trajectory model for robot manipulation that aims to improve trajectory 'stitching'—composing novel complete plans from sub-trajectories—through three components: a UNet with inpainting conditioning to enforce local temporal receptive fields, Gaussian action-noise augmentation of the training dataset, and a trajectory-splitting procedure applied at both training and inference. The authors evaluate stitching on a Franka Panda in simulation using a benchmark where boundary-condition pairs are held out, and evaluate guided planning in an obstacle-avoidance task in simulation and on real hardware. They report that Flow Planner stitches out-of-distribution boundary conditions better than MPD, transformer, and VAE baselines and can avoid obstacles up to four times larger than the baselines. Code and videos are provided.

Significance. If the results hold, the paper provides a practical recipe for making generative planners compose rather than select trajectories, with concrete architectural and dataset-augmentation choices that go beyond the usual whole-trajectory conditioning used in diffusion planners. The benchmark idea (train on opposite-side condition pairs, test on same-side pairs) is a clean way to isolate stitching ability, and the paper includes several meaningful baselines as well as a real-hardware deployment, which is a strength. However, the quantitative evidence for the headline claims is currently under-powered and some definitions are ambiguous: the stitching-error metric as written is difficult to interpret, the 'local receptive field' mechanism is asserted rather than demonstrated for the UNet used, and the obstacle-avoidance results lack statistical grounding. With a tightened evaluation and clearer reporting, this could be a useful contribution to the generative planning literature.

major comments (5)
  1. [Section 5.1, Tables 1-2] The definition of the stitching error is ambiguous and the current wording undermines the central quantitative claim. The paper states: 'We then record the mean squared error between the first and last planned states as the stitching error – since the true first and last are inpainted – and the true initial and final states.' If the endpoints are inpainted with the true states, then the endpoint error is trivially zero; if the error is instead computed at an interior point or at a stitching seam, this must be stated explicitly and the tables re-labeled accordingly. Please specify exactly which states are compared and what the conditioning inputs are, and clarify whether 'batches of 64' are independent rollouts whose standard deviations are reported.
  2. [Section 4.1 and Table 3] The paper's central explanation for the stitching improvement is that the UNet with inpainting has a 'completely local receptive field,' but this appears inconsistent with the architecture as described. The UNet in Table 3 has downsampling layers with channel dims [32, 64, 128, 256], so the effective receptive field grows with network depth, and inpainting conditions on the global start and goal states at full resolution. The claimed locality is therefore not a property of the architecture as presented. Please either provide a direct measurement of the effective receptive field (for example, via input-gradient or ablation analysis) or revise the claim to 'approximately local' and discuss the role of the bottleneck, since this mechanism is load-bearing for the paper's narrative.
  3. [Section 5.2, Fig. 5, Appendix B] The headline claim that the method 'avoid[s] obstacles up to four times as large' is not supported by the reported evidence. Fig. 5 reports only a single 'maximum obstacle radius that each method was able to reliably avoid' with no trial counts, no error bars, and no definition of 'reliably' in the main text. Appendix B defines success as reaching the goal without collision and shows success-rate curves for FP and FP+split, but it omits the MPD and VAE baselines, so the factor-of-four comparison against baselines cannot be independently checked. Please report success-rate curves with confidence intervals or error bars for all methods and state the number of rollouts per obstacle size.
  4. [Section 4.2 and Section 7 (Limitations)] The standard deviation of the Gaussian action noise is never reported. Since action-noise augmentation is claimed to be uniquely effective and the Limitations section explicitly describes a trade-off between trajectory smoothness and stitching robustness as a function of noise level, this is a key free parameter for reproducibility. Please state the value used in the experiments (in Table 3 or elsewhere) and include a sensitivity analysis showing how stitching performance and trajectory smoothness vary with the noise level.
  5. [Section 4.3 and Fig. 4] The claim that train-time trajectory splitting 'completely eliminates' mode collapse is supported only by the qualitative plots in Fig. 4a. The paper does not provide a quantitative measure of mode collapse, nor does it define the training stage or guidance scale at which the phenomenon was assessed. Please add a quantitative metric (for example, the consistency error or the variance of generated trajectories across seeds) and describe the exact conditions under which the collapse was observed and eliminated.
minor comments (8)
  1. [Section 1] There are several copyediting issues: 'plan viastitching' should be 'plan via stitching', and 'massively improvement flexibility' should be 'massively improved flexibility'.
  2. [Figure 1 caption] The caption contains an awkward comma: 'a local receptive, field' should be 'a local receptive field'.
  3. [Section 3.2] The guidance weighting bt = 1-t is described as found to work better 'in practice'; please state the range of values tried and whether this was selected on a validation set, since it is an ad hoc modification of the theoretically justified weighting.
  4. [Section 5.2] The hardware experiment is described with a single example and no quantitative success metric or number of trials. Please state how many hardware trials were performed and report success rates or at least a clear statement that this is a qualitative demonstration.
  5. [Section 5.2, Fig. 5] The sentence 'The numbers in the right figure are the maximum object radius that the arm was able to reliably avoid' should refer to 'the right panel of Fig. 5' and should state the units explicitly in the caption.
  6. [Section 7] There are two typos: 'A interesting direction' should be 'An interesting direction', and 'less destructive manor' should be 'less destructive manner'.
  7. [References] Several references are incomplete or have line-broken URLs (for example, [1], [2], [3], [9]); please ensure all entries are formatted consistently with venue and publisher information.
  8. [Abstract and Section 5] The word 'significantly' is used in the abstract and Section 5 without statistical tests. Please either add significance tests (e.g., confidence intervals or paired tests across seeds) or replace 'significantly' with a descriptive term such as 'substantially'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the core stitching and obstacle-avoidance claims rest on held-out boundary-condition tests and controlled ablations, not on fitted inputs or self-citation chains.

full rationale

The paper's central derivation is empirical rather than definitional. Flow Planner is trained on a cross-shaped dataset with held-out same-side boundary pairs, and stitching error is measured against baselines (MPD, VAE, transformers, ablation variants) under a fixed protocol. The obstacle-avoidance results compare maximum reliably avoided sphere radii under a shared analytical guidance function, and Appendix B adds success-rate curves; these are external evaluations, not quantities reconstructed from fitted parameters. The guidance weight bt=1-t is explicitly described as 'Although not theoretically justified, we found better results in practice', i.e., an ad hoc choice rather than a parameter fitted to the benchmark and then renamed a prediction. The self-citations ([24], [40]) are related-work and baseline citations; neither supplies a load-bearing uniqueness theorem or ansatz, so they do not make the argument circular. One ambiguity exists: the Section 5.1 stitching-error sentence says the true first and last states are inpainted and then compares planned endpoints with the same true states, which would be trivially zero if conditioning were exact; the nonzero reported errors show the text is incomplete, but this is a metric/protocol reporting problem, not a circular reduction, and the paper's headline claims do not depend on that single metric. Overall, the claimed improvements are supported by controlled comparisons against external baselines and by hardware deployment, so no circularity is evident.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim depends on two empirical hypotheses (local receptive field and action-noise correlation breaking) that are supported only by ablations, not by direct measurement. Standard flow matching and guidance results are imported from the literature. Several key hyperparameters (action-noise sigma, guidance scale) are chosen by hand and not all reported.

free parameters (5)
  • Action noise standard deviation = not reported
    Gaussian noise added to torques during dataset collection; central to stitching improvement, but its magnitude is not specified in the paper or appendix.
  • Train-time trajectory splitting probability = 0.5
    Half-length trajectories mixed into batches with 50% probability; chosen by hand in Section 4.3.
  • Inference-time re-noise percentage = 50%
    Initial guess re-noised up to 50% before splitting in Section 4.3.
  • Guidance scale = not reported
    Used for guided obstacle avoidance; paper says FP+split allows 'a larger guidance scale' but does not list the scale in Table 3.
  • Guidance weighting bt = bt = 1 - t
    Used instead of theoretical (1-t)/t; paper admits it is not theoretically justified (Section 3.2).
assumptions (5)
  • standard math Conditional flow matching loss gradients equal full flow matching loss gradients (Theorem in [34]).
    Invoked in Section 3.1 to justify training the velocity field with the conditional OT path.
  • standard math Classifier guidance transforms velocity fields to score functions for conditional generation.
    Used to steer generation via analytical obstacle cost in Section 3.2, citing [38].
  • domain assumption A UNet with inpainting conditioning yields a local receptive field where each state only attends to local neighbors.
    Section 4.1 states this as crucial; it is supported only indirectly by the stitching benchmark, not by measuring receptive fields.
  • domain assumption Breaking joint-state correlations via action noise is necessary for stitching.
    Hypothesized in Section 4.2; tested only by ablation, the mechanism is not directly verified.
  • domain assumption UNet architecture requires trajectory lengths that are powers of 2; arbitrary lengths cause order-of-magnitude error increases.
    Stated in Section 4.3; constrains the splitting design.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improving Trajectory Stitching with Flow Models." pith.science (2026). https://pith.science/paper/DAUZLZ4C

@misc{pith2026250507802,
  author       = {Pith},
  title        = {Pith review of: Improving Trajectory Stitching with Flow Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DAUZLZ4C}},
  note         = {Machine review of arXiv:2505.07802}
}
read the original abstract

Generative models have shown great promise as trajectory planners, given their affinity to modeling complex distributions and guidable inference process. Previous works have successfully applied these in the context of robotic manipulation but perform poorly when the required solution does not exist as a complete trajectory within the training set. We identify that this is a result of being unable to plan via stitching, and subsequently address the architectural and dataset choices needed to remedy this. On top of this, we propose a novel addition to the training and inference procedures to both stabilize and enhance these capabilities. We demonstrate the efficacy of our approach by generating plans with out of distribution boundary conditions and performing obstacle avoidance on the Franka Panda in simulation and on real hardware. In both of these tasks our method performs significantly better than the baselines and is able to avoid obstacles up to four times as large.

Figures

Figures reproduced from arXiv: 2505.07802 by the authors.

Figure 1
Figure 1. Overview: We use a flow model with three key components: a local receptive, field, dataset [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Demonstration of local receptive fields: (a) Original dataset (b) Transformer with direct [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Effect of dataset augmentation: The two figures above show the end-effector trajectories [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Trajectory splitting: (a) Training without train time trajectory splitting causes mode col [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Performance benchmark in an obstacle avoidance task on a Franka arm in simulation. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Hardware deployment, the top row shows an unguided plan and the bottom a guided one. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Collision-free success rates at different obstacle sizes [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 10 canonical work pages

  1. [1]

    Rombach, A

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer. High-resolution image syn- thesis with latent diffusion models, 2022

  2. [2]

    Podell, Z

    D. Podell, Z. English, K. Lacey, A. Blattmann, T. Dockhorn, J. M ¨uller, J. Penna, and R. Rom- bach. Sdxl: Improving latent diffusion models for high-resolution image synthesis, 2023. URL https://arxiv.org/abs/2307.01952

  3. [3]

    Esser, J

    P. Esser, J. Chiu, P. Atighehchian, J. Granskog, and A. Germanidis. Structure and content- guided video synthesis with diffusion models, 2023. URLhttps://arxiv.org/abs/2302. 03011

  4. [4]

    Janner, Y

    M. Janner, Y . Du, J. B. Tenenbaum, and S. Levine. Planning with diffusion for flexible behavior synthesis, 2022

  5. [5]

    A. Ajay, Y . Du, A. Gupta, J. Tenenbaum, T. Jaakkola, and P. Agrawal. Is conditional generative modeling all you need for decision-making?, 2023

  6. [6]

    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, 2024

  7. [7]

    J. Ho, A. Jain, and P. Abbeel. Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

  8. [8]

    Song and S

    Y . Song and S. Ermon. Generative modeling by estimating gradients of the data distribution, 2020

Show all 42 references
  1. [9]

    Esser, S

    P. Esser, S. Kulal, A. Blattmann, R. Entezari, J. M¨uller, H. Saini, Y . Levi, D. Lorenz, A. Sauer, F. Boesel, D. Podell, T. Dockhorn, Z. English, K. Lacey, A. Goodwin, Y . Marek, and R. Rom- bach. Scaling Rectified Flow Transformers for High-Resolution Image Synthesis, Mar. 2...

  2. [10]

    Polyak, A

    A. Polyak, A. Zohar, A. Brown, A. Tjandra, A. Sinha, A. Lee, A. Vyas, B. Shi, C.-Y . Ma, C.-Y . Chuang, D. Yan, D. Choudhary, D. Wang, G. Sethi, G. Pang, H. Ma, I. Misra, J. Hou, J. Wang, K. Jagadeesh, K. Li, L. Zhang, M. Singh, M. Williamson, M. Le, M. Yu, M. K. Singh, P. Zha...

  3. [11]

    M. Le, A. Vyas, B. Shi, B. Karrer, L. Sari, R. Moritz, M. Williamson, V . Manohar, Y . Adi, J. Mahadeokar, and W.-N. Hsu. V oicebox: Text-guided multilingual universal speech genera- tion at scale, 2023. URLhttps://arxiv.org/abs/2306.15687

  4. [12]

    Black, N

    K. Black, N. Brown, D. Driess, A. Esmail, M. Equi, C. Finn, N. Fusai, L. Groom, K. Haus- man, 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...

  5. [13]

    Holderrieth, M

    P. Holderrieth, M. Havasi, J. Yim, N. Shaul, I. Gat, T. Jaakkola, B. Karrer, R. T. Q. Chen, and Y . Lipman. Generator matching: Generative modeling with arbitrary markov processes, 2025. URLhttps://arxiv.org/abs/2410.20587. 10

  6. [14]

    Reuss, M

    M. Reuss, M. Li, X. Jia, and R. Lioutikov. Goal-conditioned imitation learning using score- based diffusion policies.arXiv preprint arXiv:2304.02532, 2023

  7. [15]

    Z. Wang, J. J. Hunt, and M. Zhou. Diffusion policies as an expressive policy class for offline re- inforcement learning. InThe Eleventh International Conference on Learning Representations,

  8. [16]

    Carvalho, A

    J. Carvalho, A. T. Le, M. Baierl, D. Koert, and J. Peters. Motion Planning Diffusion: Learning and Planning of Robot Motions with Diffusion Models, Mar. 2024. URLhttp://arxiv. org/abs/2308.01557. arXiv:2308.01557 [cs]

  9. [17]

    Y . Luo, C. Sun, J. B. Tenenbaum, and Y . Du. Potential based diffusion motion planning, 2024. URLhttps://arxiv.org/abs/2407.06169

  10. [18]

    W. Luo, Y . Li, R. Urtasun, and R. Zemel. Understanding the effective receptive field in deep convolutional neural networks, 2017. URLhttps://arxiv.org/abs/1701.04128

  11. [19]

    Ubukata, J

    T. Ubukata, J. Li, and K. Tei. Diffusion model for planning: A systematic literature review,

  12. [20]

    Pearce, T

    T. Pearce, T. Rashid, A. Kanervisto, D. Bignell, M. Sun, R. Georgescu, S. V . Macua, S. Z. Tan, I. Momennejad, K. Hofmann, and S. Devlin. Imitating human behaviour with diffusion models, 2023. URLhttps://arxiv.org/abs/2301.10677

  13. [21]

    H. He, C. Bai, K. Xu, Z. Yang, W. Zhang, D. Wang, B. Zhao, and X. Li. Diffusion model is an effective planner and data synthesizer for multi-task reinforcement learning, 2023. URL https://arxiv.org/abs/2305.18459

  14. [22]

    L. Wang, J. Zhao, Y . Du, E. H. Adelson, and R. Tedrake. Poco: Policy composition from and for heterogeneous robot learning, 2024. URLhttps://arxiv.org/abs/2402.02511

  15. [23]

    Huang, Y

    X. Huang, Y . Chi, R. Wang, Z. Li, X. B. Peng, S. Shao, B. Nikolic, and K. Sreenath. Diffuse- loco: Real-time legged locomotion control with diffusion from offline datasets.arXiv preprint arXiv:2404.19264, 2024

  16. [24]

    O’Mahoney, A

    R. O’Mahoney, A. L. Mitchell, W. Yu, I. Posner, and I. Havoutis. Offline adaptation of quadruped locomotion using diffusion models, 2025. URLhttps://arxiv.org/abs/2411. 08832

  17. [25]

    K. Lee, S. Kim, and J. Choi. Refining diffusion planner for reliable behavior synthesis by automatic detection of infeasible plans, 2023. URLhttps://arxiv.org/abs/2310.19427

  18. [26]

    G. Li, Y . Shan, Z. Zhu, T. Long, and W. Zhang. Diffstitch: Boosting offline reinforcement learning with diffusion-based trajectory stitching, 2024. URLhttps://arxiv.org/abs/ 2402.02439

  19. [27]

    S. Kim, Y . Choi, D. E. Matsunaga, and K.-E. Kim. Stitching sub-trajectories with conditional diffusion model for goal-conditioned offline rl, 2024. URLhttps://arxiv.org/abs/2402. 07226

  20. [28]

    J. Lee, S. Yun, T. Yun, and J. Park. Gta: Generative trajectory augmentation with guidance for offline reinforcement learning, 2024. URLhttps://arxiv.org/abs/2405.16907

  21. [29]

    Y . Luo, U. A. Mishra, Y . Du, and D. Xu. Generative trajectory stitching through diffusion composition, 2025. URLhttps://arxiv.org/abs/2503.05153

  22. [30]

    I. Char, V . Mehta, A. Villaflor, J. M. Dolan, and J. Schneider. Bats: Best action trajectory stitching, 2022. URLhttps://arxiv.org/abs/2204.12026. 11

  23. [31]

    C. A. Hepburn and G. Montana. Model-based trajectory stitching for improved offline rein- forcement learning, 2022. URLhttps://arxiv.org/abs/2211.11603

  24. [32]

    Y .-H. Wu, X. Wang, and M. Hamaya. Elastic decision transformer, 2023. URLhttps: //arxiv.org/abs/2307.02484

  25. [33]

    Lipman, M

    Y . Lipman, M. Havasi, P. Holderrieth, N. Shaul, M. Le, B. Karrer, R. T. Q. Chen, D. Lopez- Paz, H. Ben-Hamu, and I. Gat. Flow matching guide and code, 2024. URLhttps://arxiv. org/abs/2412.06264

  26. [34]

    Lipman, R

    Y . Lipman, R. T. Q. Chen, H. Ben-Hamu, M. Nickel, and M. Le. Flow matching for generative modeling, 2023. URLhttps://arxiv.org/abs/2210.02747

  27. [35]

    Nichol, P

    A. Nichol, P. Dhariwal, A. Ramesh, P. Shyam, P. Mishkin, B. McGrew, I. Sutskever, and M. Chen. Glide: Towards photorealistic image generation and editing with text-guided diffu- sion models, 2022. URLhttps://arxiv.org/abs/2112.10741

  28. [36]

    Y . Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, and B. Poole. Score-Based Generative Modeling through Stochastic Differential Equations, Feb. 2021. URLhttp:// arxiv.org/abs/2011.13456. arXiv:2011.13456 [cs, stat]

  29. [37]

    Dhariwal and A

    P. Dhariwal and A. Nichol. Diffusion Models Beat GANs on Image Synthesis, June 2021. URLhttp://arxiv.org/abs/2105.05233. arXiv:2105.05233 [cs, stat]

  30. [38]

    Zheng, M

    Q. Zheng, M. Le, N. Shaul, Y . Lipman, A. Grover, and R. T. Q. Chen. Guided flows for generative modeling and decision making, 2023. URLhttps://arxiv.org/abs/2311. 13443

  31. [39]

    Tobin, R

    J. Tobin, R. Fong, A. Ray, J. Schneider, W. Zaremba, and P. Abbeel. Domain randomization for transferring deep neural networks from simulation to the real world, 2017. URLhttps: //arxiv.org/abs/1703.06907

  32. [40]

    reliably avoid

    C.-M. Hung, S. Zhong, W. Goodwin, O. P. Jones, M. Engelcke, I. Havoutis, and I. Posner. Reaching through latent space: From joint statistics to path planning in manipulation.IEEE Robotics and Automation Letters, 7(2):5334–5341, Apr. 2022. ISSN 2377-3774. doi:10.1109/ lra.2022....

  33. [2023]

    URLhttps://openreview.net/forum?id=AHvFDPi-FA

  34. [2024]

    URLhttps://arxiv.org/abs/2408.10266

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.