pith. sign in

arxiv: 2606.31493 · v1 · pith:LM4BIDLMnew · submitted 2026-06-30 · 💻 cs.RO

ChronoFlow-Policy: Unifying Past-Current-Future Interaction Flow in Visuomotor Policy Learning

Bokai Lin , Yifu Xu , Xinyu Zhan , Hongjie Fang , Jialin Tian , Fu-Cheng Zhang , Yong-Lu Li , Cewu Lu
show 1 more author
Lixin Yang
This is my paper

Pith reviewed 2026-07-01 05:45 UTC · model grok-4.3

classification 💻 cs.RO
keywords ChronoFlowvisuomotor policydiffusion policytemporal representation3D keypointsrobot manipulationlong-horizon tasksnon-Markovian control
0
0 comments X

The pith

ChronoFlow-Policy learns one representation of past-current-future interaction dynamics to improve diffusion-based visuomotor policies.

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

The paper introduces ChronoFlow as a unified representation that encodes interaction flows across past experience, present state, and anticipated outcomes using sparse 3D keypoints of objects and the gripper. It then builds ChronoFlow-Policy, a diffusion model that co-trains this representation together with action sequences. Experiments across 14 simulated and 5 real-world manipulation tasks show consistent gains over standard diffusion baselines, with particular strength on long-horizon and non-Markovian scenarios. The core idea is that policies benefit when history and future prediction are not handled in separate modules but learned jointly inside one temporal flow.

Core claim

ChronoFlow is a temporally unified representation that captures past, current, and future interaction dynamics through sparse 3D keypoints of both objects and the gripper. ChronoFlow-Policy is a diffusion-based visuomotor policy that jointly learns ChronoFlow and action sequences through a co-training objective, producing policies that integrate historical context with future predictions rather than modeling them in isolation.

What carries the argument

ChronoFlow: a single representation of past-current-future interaction dynamics built from sparse 3D keypoints of objects and gripper, jointly trained with actions via co-training objective.

If this is right

  • Policies gain robustness on long-horizon manipulation tasks by integrating historical context and future predictions in one model.
  • Performance improves on non-Markovian scenarios where decisions depend on extended temporal context.
  • The same representation and co-training approach yields gains across both simulated and real-world manipulation settings.
  • Existing diffusion-policy baselines can be extended by adding the ChronoFlow component without changing the underlying diffusion architecture.

Where Pith is reading between the lines

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

  • The ChronoFlow approach could be tested on policy architectures other than diffusion to check whether the benefit is tied to the diffusion formulation or to the unified representation itself.
  • If the keypoint-based flow generalizes, similar sparse temporal representations might simplify policy learning in navigation or multi-agent settings where long-range timing matters.
  • A direct test would measure whether removing the future-prediction part of ChronoFlow while keeping the past-current part erodes performance on the long-horizon tasks.

Load-bearing premise

Jointly learning the ChronoFlow representation and action sequences through the co-training objective produces a genuinely unified temporal model rather than simply adding capacity that could be achieved by other means.

What would settle it

A controlled comparison in which a diffusion policy with matched parameter count but without the ChronoFlow co-training objective achieves equal performance on the same 14 simulated and 5 real tasks.

Figures

Figures reproduced from arXiv: 2606.31493 by Bokai Lin, Cewu Lu, Fu-Cheng Zhang, Hongjie Fang, Jialin Tian, Lixin Yang, Xinyu Zhan, Yifu Xu, Yong-Lu Li.

Figure 1
Figure 1. Figure 1: Motivation and Overview of ChronoFlow. Human intuition solves ma￾nipulation tasks by reasoning over past–current–future interaction: recalling how the hand and object have moved, perceiving their current relation, and anticipating how the object should move next. Inspired by this temporal reasoning, ChronoFlow pro￾vides an intermediate representation for robot that encodes unified object-gripper key￾point … view at source ↗
Figure 2
Figure 2. Figure 2: Overview of ChronoFlow-Policy. Top: Illustration of ChronoFlow: unified past-current-future object-gripper keypoint flows. Bottom: The network architecture. 3.3 ChronoFlow-Policy Based on the ChronoFlow representation, we propose ChronoFlow-Policy, a diffusion-based visuomotor policy that explicitly models interaction dynamics and leverages them for action generation. Overview. Instead of directly predicti… view at source ↗
Figure 3
Figure 3. Figure 3: Real-world manipulation tasks used for evaluation. [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Left: Predicted ChronoFlow in real-world tasks: gripper flows (top), object flows (bottom), and historical object flows (yellow/orange). Right: Ablation on Fold Towel. CFP (Full) denotes the full model. w/o CF removes ChronoFlow trajec￾tory supervision. w/o ObjPoints and w/o GriPoints remove object and gripper trajectories from ChronoFlow, respectively. w/o CF-Sampling disables stochastic subsampling of ob… view at source ↗
Figure 5
Figure 5. Figure 5: Learning curves on simulation benchmarks. [PITH_FULL_IMAGE:figures/full_fig_p024_5.png] view at source ↗
read the original abstract

Visual signals play a crucial role in policy learning by enabling models to capture object motion and interaction dynamics. Just as humans reason about actions using both past experience and anticipated outcomes, effective policies should integrate past interactions with future predictions. However, existing visuomotor policies typically model either historical context or future dynamics in isolation, lacking a unified temporal representation of interaction dynamics. In this work, we introduce \textbf{ChronoFlow}, a temporally unified representation that captures \textbf{past, current, and future} interaction dynamics through sparse 3D keypoints of both objects and the gripper. Based on this representation, we propose \textbf{ChronoFlow-Policy}, a diffusion-based visuomotor policy that jointly learns ChronoFlow and action sequences through a co-training objective. Experiments on 14 simulated tasks and 5 real-world manipulation tasks demonstrate that ChronoFlow-Policy consistently outperforms strong diffusion-policy baselines and improves robustness in long-horizon and non-Markovian manipulation scenarios.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The manuscript introduces ChronoFlow, a sparse 3D keypoint representation that unifies past, current, and future interaction dynamics of objects and the gripper, and ChronoFlow-Policy, a diffusion-based visuomotor policy that jointly optimizes the ChronoFlow representation and action sequences via a co-training objective. It reports that this approach consistently outperforms strong diffusion-policy baselines on 14 simulated tasks and 5 real-world manipulation tasks while improving robustness in long-horizon and non-Markovian scenarios.

Significance. If the performance gains can be attributed to the unified temporal structure enforced by the co-training objective rather than generic increases in capacity, the work would provide a concrete mechanism for incorporating historical context and future prediction into visuomotor policies. The evaluation across 14 simulated and 5 real-world tasks supplies a reasonably broad testbed for long-horizon manipulation.

major comments (2)
  1. [Experiments] Experiments section: the central claim that ChronoFlow plus co-training produces a unified temporal model whose structure (rather than parameter count) explains robustness gains on long-horizon/non-Markovian tasks is not isolated by any capacity-matched control. No baseline is reported that augments a diffusion policy with an auxiliary head of identical size but lacking the past-current-future flow (e.g., static keypoint regression or independent future prediction).
  2. [Method] Method (co-training objective): the joint optimization is presented as enforcing unification, yet no ablation compares the full ChronoFlow co-training loss against an equivalent-capacity variant that predicts future keypoints independently of the past-current flow; without this, the attribution of gains to the claimed unification remains open.
minor comments (2)
  1. The abstract and introduction refer to 'strong diffusion-policy baselines' without naming the specific implementations or hyper-parameters used; these details should appear in the experimental setup for reproducibility.
  2. Keypoint visualizations in the figures would benefit from explicit captions indicating which colors correspond to past, current, and future keypoints.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. The points raised about isolating the contribution of the unified temporal structure are important for strengthening the attribution of gains. We address each major comment below and will incorporate the suggested controls and ablations into the revised manuscript.

read point-by-point responses

  1. Referee: [Experiments] Experiments section: the central claim that ChronoFlow plus co-training produces a unified temporal model whose structure (rather than parameter count) explains robustness gains on long-horizon/non-Markovian tasks is not isolated by any capacity-matched control. No baseline is reported that augments a diffusion policy with an auxiliary head of identical size but lacking the past-current-future flow (e.g., static keypoint regression or independent future prediction).

    Authors: We agree that the current set of experiments lacks a capacity-matched control to isolate the effect of the unified past-current-future flow from potential capacity increases. In the revised manuscript, we will add a new baseline that augments the diffusion policy with an auxiliary head of comparable parameter count but without the temporal unification (e.g., static keypoint regression or independent future prediction). This will provide a clearer attribution of the robustness gains on long-horizon and non-Markovian tasks to the co-training objective. revision: yes

  2. Referee: [Method] Method (co-training objective): the joint optimization is presented as enforcing unification, yet no ablation compares the full ChronoFlow co-training loss against an equivalent-capacity variant that predicts future keypoints independently of the past-current flow; without this, the attribution of gains to the claimed unification remains open.

    Authors: We concur that an ablation against an equivalent-capacity variant predicting future keypoints independently would better support the claim that the joint optimization enforces unification. We will include this ablation in the revised version, training a model variant with matched capacity that decouples future keypoint prediction from the past-current flow and comparing it directly to the full ChronoFlow co-training loss. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical method with no self-referential derivations

full rationale

The paper defines ChronoFlow as a sparse 3D keypoint representation of past-current-future dynamics and proposes a diffusion policy trained via co-training objective. No equations, uniqueness theorems, or predictions are presented that reduce by construction to fitted parameters or self-citations. Outperformance is claimed via experiments on 14 simulated and 5 real tasks against diffusion baselines; the unification is an explicit design choice, not a derived result that loops back to inputs. This is a standard non-circular empirical contribution.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no information on free parameters, background axioms, or invented entities beyond the high-level description of ChronoFlow.

pith-pipeline@v0.9.1-grok · 5732 in / 1057 out tokens · 29428 ms · 2026-07-01T05:45:24.934633+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

41 extracted references · 12 canonical work pages · 5 internal anchors

  1. [1]

    In: European Conference on Computer Vision

    Bharadhwaj, H., Mottaghi, R., Gupta, A., Tulsiani, S.: Track2act: Predicting point tracks from internet videos enables generalizable robot manipulation. In: European Conference on Computer Vision. pp. 306–324. Springer (2024)

    2024

  2. [2]

    In: NeurIPS

    Chen, B., Monso, D.M., Du, Y., Simchowitz, M., Tedrake, R., Sitzmann, V.: Diffu- sion forcing: Next-token prediction meets full-sequence diffusion. In: NeurIPS. pp. 24081–24125 (2024)

    2024

  3. [3]

    History-Aware Visuomotor Policy Learning via Point Tracking.arXiv preprint arXiv:2509.17141, 2025

    Chen, J., Fang, H., Wang, C., Wang, S., Lu, C.: History-aware visuomotor policy learning via point tracking. arXiv preprint arXiv:2509.17141 (2025)

    work page arXiv 2025

  4. [4]

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

    Chen, T., Chen, Z., Chen, B., Cai, Z., Liu, Y., Li, Z., Liang, Q., Lin, X., Ge, Y., Gu, Z., et al.: Robotwin 2.0: A scalable data generator and benchmark with strong domain randomization for robust bimanual robotic manipulation. arXiv preprint arXiv:2506.18088 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  5. [5]

    In: Robotics: Science and Systems (2022)

    Eisner, B., Zhang, H., Held, D.: Flowbot3d: Learning 3d articulation flow to ma- nipulate articulated objects. In: Robotics: Science and Systems (2022)

    2022

  6. [6]

    In: IEEE International Conference on Robotics and Automation

    Fang,H.S.,Fang,H.,Tang,Z.,Liu,J.,Wang,C.,Wang,J.,Zhu,H.,Lu,C.:RH20T: A comprehensive robotic dataset for learning diverse skills in one-shot. In: IEEE International Conference on Robotics and Automation. pp. 653–660. IEEE (2024)

    2024

  7. [7]

    In: International Conference on Machine Learning (2025)

    Fang, H., Grotz, M., Pumacay, W., Wang, Y.R., Fox, D., Krishna, R., Duan, J.: Sam2act: Integrating visual foundation model with a memory architecture for robotic manipulation. In: International Conference on Machine Learning (2025)

    2025

  8. [8]

    Gkanatsios, J

    Gkanatsios, N., Xu, J., Bronars, M., Mousavian, A., Ke, T.W., Fragkiadaki, K.: 3d flowmatch actor: Unified 3d policy for single-and dual-arm manipulation. arXiv preprint arXiv:2508.11002 (2025)

    work page arXiv 2025

  9. [9]

    In: IEEE International Conference on Robotics and Automation

    Hsu, C.C., Wen, B., Xu, J., Narang, Y., Wang, X., Zhu, Y., Biswas, J., Birchfield, S.: Spot: Se(3) pose trajectory diffusion for object-centric manipulation. In: IEEE International Conference on Robotics and Automation. pp. 4853–4860 (2025)

    2025

  10. [10]

    In: International Conference on Machine Learning

    Hu, Y., Guo, Y., Wang, P., Chen, X., Wang, Y.J., Zhang, J., Sreenath, K., Lu, C., Chen, J.: Video prediction policy: A generalist robot policy with predictive visual representations. In: International Conference on Machine Learning. pp. 24328– 24346 (2025)

    2025

  11. [11]

    Huang, Y ., Zhang, J., Zou, S., Liu, X., Hu, R., and Xu, K

    Huang, W., Chao, Y.W., Mousavian, A., Liu, M.Y., Fox, D., Mo, K., Fei-Fei, L.: Pointworld: Scaling 3d world models for in-the-wild robotic manipulation. arXiv preprint arXiv:2601.03782 (2026)

    work page arXiv 2026

  12. [12]

    In: European conference on computer vision

    Huang, Z., Shi, X., Zhang, C., Wang, Q., Cheung, K.C., Qin, H., Dai, J., Li, H.: Flowformer: A transformer architecture for optical flow. In: European conference on computer vision. pp. 668–685. Springer (2022)

    2022

  13. [13]

    In: International Conference on Machine Learning

    Jin, Y., Sun, Z., Xu, K., Chen, L., Jiang, H., Huang, Q., Song, C., Liu, Y., Zhang, D., Song, Y., et al.: Video-lavit: unified video-language pre-training with decoupled visual-motional tokenization. In: International Conference on Machine Learning. pp. 22185–22209 (2024)

    2024

  14. [14]

    arXiv preprint arXiv:2506.19816 (2025)

    Li, H., Yang, S., Chen, Y., Chen, X., Yang, X., Tian, Y., Wang, H., Wang, T., Lin, D., Zhao, F., et al.: Cronusvla: Towards efficient and robust manipulation via multi-frame vision-language-action modeling. arXiv preprint arXiv:2506.19816 (2025)

    work page arXiv 2025

  15. [15]

    Causal World Modeling for Robot Control

    Li, L., Zhang, Q., Luo, Y., Yang, S., Wang, R., Han, F., Yu, M., Gao, Z., Xue, N., Zhu, X., Shen, Y., Xu, Y.: Causal world modeling for robot control. arXiv preprint arXiv:2601.21998 (2026) ChronoFlow-Policy 17

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  16. [16]

    arXiv preprint arXiv:2509.18676 (2025)

    Noh, S., Nam, D., Kim, K., Lee, G., Yu, Y., Kang, R., Lee, K.: 3d flow diffusion policy: Visuomotor policy learning via generating flow in 3d space. arXiv preprint arXiv:2509.18676 (2025)

    work page arXiv 2025

  17. [17]

    mimic-video: Video-Action Models for Generalizable Robot Control Beyond VLAs

    Pai, J., Achenbach, L., Montesinos, V., Forrai, B., Mees, O., Nava, E.: mimic-video: Video-action models for generalizable robot control beyond vlas. arXiv preprint arXiv:2512.15692 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  18. [18]

    Peebles,W.,Xie,S.:Scalablediffusionmodelswithtransformers.In:Proceedingsof the IEEE/CVF international conference on computer vision. pp. 4195–4205 (2023)

    2023

  19. [19]

    In: International Conference on Learning Representations (2025)

    Ravi, N., Gabeur, V., Hu, Y.T., Hu, R., Ryali, C., Ma, T., Khedr, H., Rädle, R., Rolland, C., Gustafson, L., et al.: Sam 2: Segment anything in images and videos. In: International Conference on Learning Representations (2025)

    2025

  20. [20]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomedi- cal image segmentation. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 234–241. Springer (2015)

    2015

  21. [22]

    MemoryVLA: Perceptual-Cognitive Memory in Vision-Language-Action Models for Robotic Manipulation

    Shi, H., Xie, B., Liu, Y., Sun, L., Liu, F., Wang, T., Zhou, E., Fan, H., Zhang, X., Huang, G.: Memoryvla: Perceptual-cognitive memory in vision-language-action models for robotic manipulation. arXiv preprint arXiv:2508.19236 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  22. [23]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Shi, X., Huang, Z., Li, D., Zhang, M., Cheung, K.C., See, S., Qin, H., Dai, J., Li, H.: Flowformer++: Masked cost volume autoencoding for pretraining optical flow estimation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 1599–1610 (2023)

    2023

  23. [24]

    Freqpolicy: Efficient flow-based visuomotor policy via frequency consistency, 2025

    Su, Y., Liu, N., Chen, D., Zhao, Z., Wu, K., Li, M., Xu, Z., Che, Z., Tang, J.: Freqpolicy: Efficient flow-based visuomotor policy via frequency consistency. arXiv preprint arXiv:2506.08822 (2025)

    work page arXiv 2025

  24. [25]

    IEEE Robotics and Automation Letters 10(7), 7428–7435 (2025)

    Su, Y., Zhan, X., Fang, H., Li, Y., Lu, C., Yang, L.: Motion before action: Diffusing object motion as manipulation condition. IEEE Robotics and Automation Letters 10(7), 7428–7435 (2025)

    2025

  25. [26]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Tang, X., Qiu, J., Xie, L., Tian, Y., Jiao, J., Ye, Q.: Adaptive keyframe sampling for long video understanding. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 29118–29128 (2025)

    2025

  26. [27]

    Torne,M.,Pertsch,K.,Walke,H.,Vedder,K.,Nair,S.,Ichter,B.,Ren,A.Z.,Wang, H., Tang, J., Stachowicz, K., Dhabalia, K., Equi, M., Vuong, Q., Springenberg, J.T., Levine, S., Finn, C., Driess, D.: Mem: Multi-scale embodied memory for vision language action models (2026)

    2026

  27. [28]

    In: CoRL (2025)

    Torne, M., Tang, A., Liu, Y., Finn, C.: Learning long-context diffusion policies via past-token prediction. In: CoRL (2025)

    2025

  28. [29]

    In: IEEE/RSJ International Conference on Intelligent Robots and Systems

    Wang, C., Fang, H., Fang, H.S., Lu, C.: Rise: 3d perception makes real-world robot imitation simple and effective. In: IEEE/RSJ International Conference on Intelligent Robots and Systems. pp. 2870–2877 (2024)

    2024

  29. [30]

    In: European Conference on Computer Vision (ECCV) (2026)

    Wang, X., Wang, C., Xu, Y., Ye, M., Zhang, F., Tian, J., Zhan, X., Zhu, L., Lu, C., Yang, L.: LaMP: Learning vision-language-action policy with 3d scene flow as latent motion prior. In: European Conference on Computer Vision (ECCV) (2026)

    2026

  30. [31]

    In: Robotics: Science and Systems (2024)

    Wen, C., Lin, X., So, J.I.R., Chen, K., Dou, Q., Gao, Y., Abbeel, P.: Any-point trajectory modeling for policy learning. In: Robotics: Science and Systems (2024)

    2024

  31. [32]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Xiao, Y., Wang, J., Xue, N., Karaev, N., Makarov, Y., Kang, B., Zhu, X., Bao, H., Shen, Y., Zhou, X.: Spatialtrackerv2: Advancing 3d point tracking with explicit camera motion. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 6726–6737 (2025) 18 B. Lin et al

    2025

  32. [33]

    In: Conference on Robot Learning

    Xu, M., Xu, Z., Xu, Y., Chi, C., Wetzstein, G., Veloso, M., Song, S.: Flow as the cross-domain manipulation interface. In: Conference on Robot Learning. vol. 270, pp. 2475–2499. PMLR (2024)

    2024

  33. [34]

    In: Conference on Language Modeling (2025)

    Xu, M., Gao, M., Li, S., Lu, J., Gan, Z., Lai, Z., Cao, M., Kang, K., Yang, Y., Dehghan, A.: Slowfast-llava-1.5: A family of token-efficient video large language models for long-form video understanding. In: Conference on Language Modeling (2025)

    2025

  34. [35]

    World Action Models are Zero-shot Policies

    Ye, S., Ge, Y., Zheng, K., Gao, S., Yu, S., Kurian, G., Indupuru, S., Tan, Y.L., Zhu, C., Xiang, J., et al.: World action models are zero-shot policies. arXiv preprint arXiv:2602.15922 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  35. [36]

    In: International Conference on Learning Representations (2025)

    Yu, S., Jin, C., Wang, H., Chen, Z., Jin, S., Zuo, Z., Xu, X., Sun, Z., Zhang, B., Wu, J., et al.: Frame-voyager: Learning to query frames for video large language models. In: International Conference on Learning Representations (2025)

    2025

  36. [37]

    In: Conference on Robot Learning

    Yu, T., Quillen, D., He, Z., Julian, R., Hausman, K., Finn, C., Levine, S.: Meta- world:Abenchmarkandevaluationformulti-taskandmetareinforcementlearning. In: Conference on Robot Learning. Proceedings of Machine Learning Research, vol. 100, pp. 1094–1100. PMLR (2019)

    2019

  37. [38]

    In: Robotics: Science and Systems (2024)

    Ze, Y., Zhang, G., Zhang, K., Hu, C., Wang, M., Xu, H.: 3d diffusion policy: Gen- eralizable visuomotor policy learning via simple 3d representations. In: Robotics: Science and Systems (2024)

    2024

  38. [39]

    arXiv preprint arXiv:2504.14717 (2025)

    Zhang, B., Ke, L., Harley, A.W., Fragkiadaki, K.: Tapip3d: Tracking any point in persistent 3d geometry. arXiv preprint arXiv:2504.14717 (2025)

    work page arXiv 2025

  39. [40]

    In: CVPR (2026)

    Zhang, C., Le Moing, G., Koppula, S., Rocco, I., Momeni, L., Xie, J., Sun, S., Sukthankar, R., Barral, J.K., Hadsell, R., Ghahramani, Z., Zisserman, A., Zhang, J., Sajjadi, M.S.M.: Efficiently reconstructing dynamic scenes one d4rt at a time. In: CVPR (2026)

    2026

  40. [41]

    In: International Conference on Learning Represen- tations (2025)

    Zheng, R., Liang, Y., Huang, S., Gao, J., Daumé III, H., Kolobov, A., Huang, F., Yang, J.: Tracevla: Visual trace prompting enhances spatial-temporal awareness for generalist robotic policies. In: International Conference on Learning Represen- tations (2025)

    2025

  41. [42]

    Zhu, C., Yu, R., Feng, S., Burchfiel, B., Shah, P., Gupta, A.: Unified world models: Coupling video and action diffusion for pretraining on large robotic datasets. In: Proceedings of Robotics: Science and Systems (RSS) (2025) Appendix A ChronoFlow Implementation Details This section details the construction ofChronoFlow, including how gripper and object k...

    2025