pith. sign in

arxiv: 2606.20562 · v1 · pith:NVUP7W2Snew · submitted 2026-06-18 · 💻 cs.RO

MemoryWAM: Efficient World Action Modeling with Persistent Memory

Sizhe Yang , Juncheng Mu , Tianming Wei , Chenhao Lu , Xiaofan Li , Linning Xu , Zhengrong Xue , Zhecheng Yuan
show 3 more authors
Dahua Lin Jiangmiao Pang Huazhe Xu
This is my paper

Pith reviewed 2026-06-26 16:58 UTC · model grok-4.3

classification 💻 cs.RO
keywords robotic manipulationworld action modelspersistent memoryhybrid memorylong-horizon tasksvision-language-actionefficient inference
0
0 comments X

The pith

MemoryWAM equips world action models with a hybrid persistent memory of recent frames, event anchors, and gist tokens to handle long-horizon robotic tasks efficiently.

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

The paper targets the core limitation in existing world action models where efficient versions can only look at a short recent window and therefore fail when past observations matter, while versions that keep full history pay growing time and memory costs. MemoryWAM replaces that choice with a fixed-size hybrid store: the last few frames for detail, selected past frames at event boundaries for structure, and a small set of gist tokens that compress everything earlier. A custom attention pattern then lets the model pull either the detailed short window or the compressed long context as needed. This design is shown to improve success rates on memory-heavy manipulation sequences in both simulation and real robots without the usual increase in inference cost. A reader cares because most practical robot tasks, such as rearranging objects whose locations were observed minutes ago, are non-Markovian and therefore expose exactly this trade-off.

Core claim

MemoryWAM uses a hybrid memory design that combines recent frames, event-boundary anchor frames, and compact gist tokens that summarize long-range history. A tailored attention mechanism enables retrieval of both detailed short-term context and compressed long-term context, supporting memory-dependent decision-making with reduced inference latency and GPU memory usage. Across long-horizon, memory-dependent manipulation tasks in both simulation and the real world, MemoryWAM outperforms strong vision-language-action and WAM baselines while maintaining favorable computational efficiency.

What carries the argument

Hybrid memory of recent frames plus event-boundary anchor frames plus compact gist tokens, retrieved via a tailored attention mechanism that separates short-term detail from long-term compression.

If this is right

  • The same memory structure can be used on any long-horizon task whose required history fits inside the gist-token budget.
  • Inference cost stays roughly constant even as the total number of past observations grows.
  • Real-robot deployment becomes practical for tasks that previously forced either short memory or slow full-history models.
  • The approach is compatible with existing vision-language-action pipelines by swapping only the memory module.

Where Pith is reading between the lines

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

  • The same hybrid store could be tested on non-robotics sequential problems such as long video understanding where only certain events matter.
  • If gist tokens prove too lossy on some tasks, the design suggests a natural next step of making the number of gist tokens task-adaptive rather than fixed.
  • Real-world success already reported implies the method tolerates sensor noise better than pure long-context transformers that overfit to exact pixel history.

Load-bearing premise

The chosen mix of recent frames, event-boundary anchors, and gist tokens plus the tailored attention is enough to recover every fact the robot needs for its decisions without losing anything critical.

What would settle it

A long sequence in which success requires recalling a visual detail from before the first anchor frame that the gist tokens also fail to preserve, causing MemoryWAM to choose a wrong action while a full-history model succeeds.

Figures

Figures reproduced from arXiv: 2606.20562 by Chenhao Lu, Dahua Lin, Huazhe Xu, Jiangmiao Pang, Juncheng Mu, Linning Xu, Sizhe Yang, Tianming Wei, Xiaofan Li, Zhecheng Yuan, Zhengrong Xue.

Figure 1
Figure 1. Figure 1: Overview. Prior WAMs typically face a memory-efficiency trade-off: sliding-window memory is efficient but forgets long-range context, while full-history KV caching preserves context but scales linearly with trajectory length N. MemoryWAM instead introduces hybrid memory: recent frames for short-term memory, anchor frames for event-boundary context, and gist tokens for long-range history. This reduces both … view at source ↗
Figure 2
Figure 2. Figure 2: MemoryWAM adopts an MoT architecture with a video DiT and an action DiT. Video [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Attention mask of MemoryWAM. Ex￾ample with three frames, one anchor frame, and one recent frame. f denotes clean video frames, g indicates gist tokens, and a represents actions to be denoised. The video frames to be denoised are omitted, as they and the actions attend to the same historical context. Gist memory. While short-term and event￾boundary memories preserve selected frames in their entirety, they c… view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of memory mechanisms. We compare full attention, test-time training (TTT), [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of the real-world tasks. Our hardware platform consists of an ARX dual-arm robot and a RealSense D455 camera that provides RGB observa￾tions. We compare MemoryWAM with two representative baselines: π0.5 [62] and LingBot-VA [7]. We design two chal￾lenging memory-dependent tasks, Shell Game and Look and Press, as shown in [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Hardware Setup. The experimental platform comprises a dual-arm robotic system equipped with RealSense D455 cameras for vi￾sual perception of the workspace. Hardware Setup. The robotic platform used in the experiments consists of an ARX dual-arm robot, with each arm equipped with a parallel gripper. A RealSense D455 camera captures RGB images of the workspace. The complete hardware setup is illustrated in … view at source ↗
read the original abstract

Robust robotic manipulation in the real world requires not only an understanding of the current observation, but also memory and dynamics modeling. World action models (WAMs) possess these capabilities by jointly modeling visual foresight and actions conditioned on both current and historical observations, making them a promising paradigm for robotic manipulation. However, existing WAMs face a fundamental trade-off: methods with efficient inference typically condition only on a bounded window of recent observations and therefore struggle in non-Markovian environments, whereas methods that preserve long histories incur time and space costs that grow substantially with sequence length. To address this challenge, we introduce MemoryWAM, a world action model with efficient persistent memory. MemoryWAM uses a hybrid memory design that combines recent frames, event-boundary anchor frames, and compact gist tokens that summarize long-range history. A tailored attention mechanism enables retrieval of both detailed short-term context and compressed long-term context, supporting memory-dependent decision-making with reduced inference latency and GPU memory usage. Across long-horizon, memory-dependent manipulation tasks in both simulation and the real world, MemoryWAM outperforms strong vision-language-action (VLA) and WAM baselines while maintaining favorable computational efficiency.

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

1 major / 0 minor

Summary. The manuscript introduces MemoryWAM, a world action model with persistent memory for robotic manipulation. It proposes a hybrid memory architecture combining recent frames, event-boundary anchor frames, and compact gist tokens, paired with a tailored attention mechanism, to jointly model visual foresight and actions while avoiding the efficiency penalties of full history conditioning. The central empirical claim is that this design enables outperformance over strong VLA and WAM baselines on long-horizon, memory-dependent tasks in both simulation and the real world while preserving favorable inference latency and GPU memory usage.

Significance. If the experimental results hold, the work would be significant because it directly targets the core efficiency-memory trade-off in world action models, a recognized limitation when operating in non-Markovian environments. The hybrid memory construction offers a concrete, implementable route to long-range context retrieval without linear growth in cost, which could improve robustness in real-world manipulation. The paper explicitly positions its experiments as a test of whether the proposed components suffice for memory-dependent decisions.

major comments (1)
  1. [Abstract] Abstract: the central claim of outperformance and efficiency gains is asserted without any quantitative results, baseline descriptions, error bars, dataset details, or experimental protocol. This absence prevents evaluation of the empirical contribution that the hybrid memory design is sufficient for the stated tasks.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback and positive assessment of the work's significance. We address the single major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of outperformance and efficiency gains is asserted without any quantitative results, baseline descriptions, error bars, dataset details, or experimental protocol. This absence prevents evaluation of the empirical contribution that the hybrid memory design is sufficient for the stated tasks.

    Authors: We agree that the abstract would benefit from including key quantitative highlights to better support the claims and facilitate evaluation. In the revised version, we will update the abstract to concisely report representative results (e.g., success rates on memory-dependent tasks, baseline comparisons, and efficiency metrics) drawn from the experiments detailed in Sections 4 and 5, while preserving the abstract's length constraints. The main text already contains full experimental protocols, error bars, and dataset descriptions. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper introduces an empirical architecture (hybrid memory with recent frames, event-boundary anchors, gist tokens, and tailored attention) for world action modeling and evaluates it on long-horizon robotic tasks. No equations, derivations, or first-principles claims are present that reduce by construction to fitted inputs, self-definitions, or self-citation chains. The central claim is an empirical performance statement tested via experiments, not a mathematical reduction. Any self-citations are incidental and non-load-bearing for the architecture or results.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 2 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities beyond the high-level memory components named in the text.

invented entities (2)
  • event-boundary anchor frames no independent evidence
    purpose: Mark key moments for long-range context
    Named component of the hybrid memory design
  • compact gist tokens no independent evidence
    purpose: Summarize long-range history in compressed form
    Named component of the hybrid memory design

pith-pipeline@v0.9.1-grok · 5771 in / 1144 out tokens · 19497 ms · 2026-06-26T16:58:47.227999+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

68 extracted references · 1 canonical work pages

  1. [1]

    T. Chen, Y . Wang, M. Li, Y . Qin, H. Shi, Z. Li, Y . Hu, Y . Zhang, K. Wang, Y . Chen, et al. Rmbench: Memory-dependent robotic manipulation benchmark with insights into policy design. arXiv preprint arXiv:2603.01229, 2026

    arXiv 2026

  2. [2]

    Brohan, N

    A. Brohan, N. Brown, J. Carbajal, et al. Rt-2: Vision-language-action models transfer web knowledge to robotic control.arXiv preprint arXiv:2307.15818, 2023

    Pith/arXiv arXiv 2023

  3. [3]

    M. J. Kim, K. Pertsch, S. Karamcheti, T. Xiao, A. Balakrishna, S. Nair, et al. Openvla: An open-source vision-language-action model.arXiv preprint arXiv:2406.09246, 2024

    Pith/arXiv arXiv 2024

  4. [4]

    Black, N

    K. Black, N. Brown, D. Driess, A. Esmail, M. Equi, C. Finn, et al. π0: A vision-language-action flow model for general robot control.arXiv preprint arXiv:2410.24164, 2024

    Pith/arXiv arXiv 2024

  5. [5]

    Bjorck, F

    J. Bjorck, F. Castañeda, N. Cherniadev, X. Da, R. Ding, L. Fan, Y . Fang, D. Fox, F. Hu, S. Huang, et al. Gr00t n1: An open foundation model for generalist humanoid robots.arXiv preprint arXiv:2503.14734, 2025

    Pith/arXiv arXiv 2025

  6. [6]

    S. Liu, L. Wu, B. Li, H. Tan, H. Chen, Z. Wang, K. Xu, H. Su, and J. Zhu. Rdt-1b: A diffusion foundation model for bimanual manipulation.arXiv preprint arXiv:2410.07864, 2024

    Pith/arXiv arXiv 2024

  7. [7]

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

    Pith/arXiv arXiv 2026

  8. [8]

    H. Luo, W. Zhang, Y . Feng, S. Zheng, H. Xu, C. Xu, Z. Xi, Y . Fu, and Z. Lu. Being-h0.7: A latent world-action model from egocentric videos.arXiv preprint arXiv:2605.00078, 2026

    Pith/arXiv arXiv 2026

  9. [9]

    R. A. Team. Causal video models are data-efficient robot policy learners.Rhoda AI Blog, 2026

    2026

  10. [10]

    H. Bi, H. Tan, S. Xie, Z. Wang, S. Huang, H. Liu, et al. Motus: A unified latent action world model.arXiv preprint arXiv:2512.13030, 2025

    Pith/arXiv arXiv 2025

  11. [11]

    S. Ye, Y . Ge, K. Zheng, S. Gao, S. Yu, G. Kurian, et al. World action models are zero-shot policies.arXiv preprint arXiv:2602.15922, 2026

    Pith/arXiv arXiv 2026

  12. [12]

    M. J. Kim, Y . Gao, T.-Y . Lin, Y .-C. Lin, Y . Ge, G. Lam, P. Liang, S. Song, M.-Y . Liu, C. Finn, and J. Gu. Cosmos policy: Fine-tuning video models for visuomotor control and planning. arXiv preprint arXiv:2601.16163, 2026

    Pith/arXiv arXiv 2026

  13. [13]

    T. Ma, J. Zheng, Z. Wang, C. Jiang, A. Cui, J. Liang, and S. Yang. Dit4dit: Jointly modeling video dynamics and actions for generalizable robot control.arXiv preprint arXiv:2603.10448, 2026

    arXiv 2026

  14. [14]

    T. Yuan, Z. Dong, Y . Liu, and H. Zhao. Fast-wam: Do world action models need test-time future imagination?arXiv preprint arXiv:2603.16666, 2026

    Pith/arXiv arXiv 2026

  15. [15]

    A. Ye, B. Wang, C. Ni, G. Huang, G. Zhao, H. Li, et al. Gigaworld-policy: An efficient action-centered world–action model.arXiv preprint arXiv:2603.17240, 2026

    arXiv 2026

  16. [16]

    J. Guo, Q. Li, P. Li, Z. Chen, N. Sun, Y . Su, H. Wang, Y . Zhang, X. Li, and H. Liu. Unified 4d world action modeling from video priors with asynchronous denoising.arXiv preprint arXiv:2604.26694, 2026

    Pith/arXiv arXiv 2026

  17. [17]

    M. Team, C. Xiang, F. Bao, H. Liu, H. Tan, H. Bi, J. Li, J. Liu, J. Pang, K. Jing, et al. Motubrain: An advanced world action model for robot control.arXiv preprint arXiv:2604.27792, 2026

    Pith/arXiv arXiv 2026

  18. [18]

    R. C. Atkinson and R. M. Shiffrin. Human memory: A proposed system and its control processes. InThe Psychology of Learning and Motivation, volume 2, pages 89–195. Academic Press, 1968. 10

    1968

  19. [19]

    A. D. Baddeley and G. Hitch. Working memory. InPsychology of Learning and Motivation, volume 8, pages 47–89. Academic Press, 1974

    1974

  20. [20]

    C. J. Brainerd and V . F. Reyna.The Science of False Memory. Oxford University Press, 2005

    2005

  21. [21]

    J. M. Zacks, N. K. Speer, K. M. Swallow, T. S. Braver, and J. R. Reynolds. Event perception: A mind-brain perspective.Psychological Bulletin, 133(2):273–293, 2007

    2007

  22. [22]

    Ghosh, H

    Octo Model Team, D. Ghosh, H. Walke, K. Pertsch, K. Black, et al. Octo: An open-source generalist robot policy.arXiv preprint arXiv:2405.12213, 2024

    Pith/arXiv arXiv 2024

  23. [23]

    Galaxea g0: Open-world dataset and dual-system vision-language-action model

    Galaxea Team. Galaxea g0: Open-world dataset and dual-system vision-language-action model. arXiv preprint arXiv:2509.00576, 2025

    arXiv 2025

  24. [24]

    P. Li, Y . Chen, H. Wu, X. Ma, X. Wu, Y . Huang, L. Wang, T. Kong, and T. Tan. BridgeVLA: Input-output alignment for efficient 3d manipulation learning with vision-language models. arXiv preprint arXiv:2506.07961, 2025

    arXiv 2025

  25. [25]

    D. Qu, H. Song, Q. Chen, Y . Yao, X. Ye, Y . Ding, Z. Wang, J. Gu, B. Zhao, D. Wang, and X. Li. SpatialVLA: Exploring spatial representations for visual-language-action model.arXiv preprint arXiv:2501.15830, 2025

    Pith/arXiv arXiv 2025

  26. [26]

    J. Wen, Y . Zhu, J. Li, Z. Tang, C. Shen, and F. Feng. Dexvla: Vision-language model with plug-in diffusion expert for general robot control.arXiv preprint arXiv:2502.05855, 2025

    Pith/arXiv arXiv 2025

  27. [27]

    O. X.-E. Collaboration. Open x-embodiment: Robotic learning datasets and rt-x models.arXiv preprint arXiv:2310.08864, 2023

    Pith/arXiv arXiv 2023

  28. [28]

    Khazatsky and et al

    A. Khazatsky and et al. Droid: A large-scale in-the-wild robot manipulation dataset.arXiv preprint arXiv:2403.12945, 2024

    Pith/arXiv arXiv 2024

  29. [29]

    Zhang, Z

    Z. Zhang, Z. Li, B. Rahmati, R. H. Yang, Y . Ma, et al. Do world action models generalize better than vlas? a robustness study.arXiv preprint arXiv:2603.22078, 2026

    Pith/arXiv arXiv 2026

  30. [30]

    Y . Du, M. Yang, B. Dai, H. Dai, O. Nachum, J. B. Tenenbaum, D. Schuurmans, and P. Abbeel. Learning universal policies via text-guided video generation. InAdvances in Neural Information Processing Systems, 2023

    2023

  31. [31]

    Y . Feng, H. Tan, X. Mao, G. Liu, S. Huang, C. Xiang, H. Su, and J. Zhu. Vidar: Embodied video diffusion model for generalist bimanual manipulation.arXiv preprint arXiv:2507.12898, 2025

    Pith/arXiv arXiv 2025

  32. [32]

    Bharadhwaj, D

    H. Bharadhwaj, D. Dwibedi, A. Gupta, S. Tulsiani, C. Doersch, T. Xiao, D. Shah, F. Xia, D. Sadigh, and S. Kirmani. Gen2act: Human video generation in novel scenarios enables generalizable robot manipulation.arXiv preprint arXiv:2409.16283, 2024

    Pith/arXiv arXiv 2024

  33. [33]

    S. Zhou, Y . Du, J. Chen, Y . Li, D.-Y . Yeung, and C. Gan. Robodreamer: Learning compositional world models for robot imagination.arXiv preprint arXiv:2404.12377, 2024

    Pith/arXiv arXiv 2024

  34. [34]

    Y . Hu, Y . Guo, P. Wang, X. Chen, Y .-J. Wang, J. Zhang, K. Sreenath, C. Lu, and J. Chen. Video prediction policy: A generalist robot policy with predictive visual representations.arXiv preprint arXiv:2412.14803, 2024

    Pith/arXiv arXiv 2024

  35. [35]

    Y . Tian, S. Yang, J. Zeng, P. Wang, D. Lin, H. Dong, and J. Pang. Predictive inverse dynamics models are scalable learners for robotic manipulation.arXiv preprint arXiv:2412.15109, 2024

    Pith/arXiv arXiv 2024

  36. [36]

    J. Pai, L. Achenbach, V . Montesinos, B. Forrai, O. Mees, and E. Nava. mimic-video: Video- action models for generalizable robot control beyond vlas.arXiv preprint arXiv:2512.15692, 2025. 11

    Pith/arXiv arXiv 2025

  37. [37]

    Liang, P

    J. Liang, P. Tokmakov, R. Liu, S. Sudhakar, P. Shah, R. Ambrus, and C. V ondrick. Video generators are robot policies.arXiv preprint arXiv:2508.00795, 2025

    Pith/arXiv arXiv 2025

  38. [38]

    Y . Su, S. Chen, H. Shi, M. Liu, Z. Zhang, N. Huang, W. Zhong, Z. Zhu, Y . Liu, and X. Liu. World guidance: World modeling in condition space for action generation.arXiv preprint arXiv:2602.22010, 2026

    arXiv 2026

  39. [39]

    Cheang, G

    C.-L. Cheang, G. Chen, Y . Jing, T. Kong, H. Li, Y . Li, Y . Liu, H. Wu, J. Xu, Y . Yang, H. Zhang, and M. Zhu. Gr-2: A generative video-language-action model with web-scale knowledge for robot manipulation.arXiv preprint arXiv:2410.06158, 2024

    Pith/arXiv arXiv 2024

  40. [40]

    S. Li, V . Yao, C. Yang, T. Qu, R. Cheng, R. Yu, H. Lu, N. V on, V . Chen, Y . Tang, et al. Wall-wm: Carving world action modeling at the event joints.arXiv preprint arXiv:2606.01955, 2026

    Pith/arXiv arXiv 2026

  41. [41]

    J. L. Elman. Finding structure in time.Cognitive Science, 14(2):179–211, 1990

    1990

  42. [42]

    Neural Computation , volume =

    S. Hochreiter and J. Schmidhuber. Long short-term memory.Neural Computation, 9(8): 1735–1780, 1997. doi:10.1162/neco.1997.9.8.1735

    work page doi:10.1162/neco.1997.9.8.1735 1997

  43. [43]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin. Attention is all you need. InAdvances in Neural Information Processing Systems, 2017

    2017

  44. [44]

    Katharopoulos, A

    A. Katharopoulos, A. Vyas, N. Pappas, and F. Fleuret. Transformers are RNNs: Fast autore- gressive transformers with linear attention. InProceedings of the International Conference on Machine Learning, 2020

    2020

  45. [45]

    Schlag, K

    I. Schlag, K. Irie, and J. Schmidhuber. Linear transformers are secretly fast weight programmers. InProceedings of the International Conference on Machine Learning, 2021

    2021

  46. [46]

    S. Yang, J. Kautz, and A. Hatamizadeh. Gated delta networks: Improving mamba2 with delta rule. InInternational Conference on Learning Representations, volume 2025, pages 29687–29707, 2025

    2025

  47. [47]

    Y . Sun, X. Li, K. Dalal, J. Xu, A. Vikram, G. Zhang, Y . Dubois, X. Chen, X. Wang, S. Koyejo, T. Hashimoto, and C. Guestrin. Learning to (learn at test time): RNNs with expressive hidden states. InProceedings of the International Conference on Machine Learning, 2025

    2025

  48. [48]

    Zhang, C

    J. Zhang, C. Herrmann, J. Hur, C. Sun, M.-H. Yang, F. Cole, T. Darrell, and D. Sun. LoGeR: Long-context geometric reconstruction with hybrid memory.arXiv preprint arXiv:2603.03269, 2026

    Pith/arXiv arXiv 2026

  49. [49]

    T. Xie, P. Yang, Y . Jin, Y . Cai, W. Yin, W. Ren, Q. Zhang, W. Hua, S. Peng, X. Guo, and X. Zhou. Scal3R: Scalable test-time training for large-scale 3d reconstruction.arXiv preprint arXiv:2604.08542, 2026

    Pith/arXiv arXiv 2026

  50. [50]

    L.-Z. Chen, J. Gao, Y . Chen, K. L. Cheng, Y . Sun, L. Hu, N. Xue, X. Zhu, Y . Shen, Y . Yao, and Y . Xu. Geometric context transformer for streaming 3d reconstruction.arXiv preprint arXiv:2604.14141, 2026

    Pith/arXiv arXiv 2026

  51. [51]

    Zhang, S

    L. Zhang, S. Cai, M. Li, G. Wetzstein, and M. Agrawala. Frame context packing and drift prevention in next-frame-prediction video diffusion models.Advances in Neural Information Processing Systems, 38:30546–30566, 2026

    2026

  52. [52]

    J. Yu, J. Bai, Y . Qin, Q. Liu, X. Wang, P. Wan, D. Zhang, and X. Liu. Context as memory: Scene-consistent interactive long video generation with memory retrieval. InProceedings of the SIGGRAPH Asia 2025 Conference Papers, pages 1–11, 2025. 12

    2025

  53. [53]

    Dalal, D

    K. Dalal, D. Koceja, J. Xu, Y . Zhao, S. Han, K. C. Cheung, J. Kautz, Y . Choi, Y . Sun, and X. Wang. One-minute video generation with test-time training. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 17702–17711, 2025

    2025

  54. [54]

    Zhang, S

    T. Zhang, S. Bi, Y . Hong, K. Zhang, F. Luan, S. Yang, K. Sunkavalli, W. T. Freeman, and H. Tan. Test-time training done right.arXiv preprint arXiv:2505.23884, 2025

    Pith/arXiv arXiv 2025

  55. [55]

    Torne, K

    M. Torne, K. Pertsch, H. Walke, K. Vedder, S. Nair, D. Driess, et al. Mem: Multi-scale embodied memory for vision language action models.arXiv preprint arXiv:2603.03596, 2026

    arXiv 2026

  56. [56]

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

    Pith/arXiv arXiv 2025

  57. [57]

    H. Li, F. Shen, D. Chen, L. Yang, X. Wang, J. Shi, Z. Bing, Z. Liu, and A. Knoll. Remem-vla: Empowering vision-language-action model with memory via dual-level recurrent queries.arXiv preprint arXiv:2603.12942, 2026

    arXiv 2026

  58. [58]

    Sridhar, J

    A. Sridhar, J. Pan, S. Sharma, and C. Finn. Memer: Scaling up memory for robot control via experience retrieval.arXiv preprint arXiv:2510.20328, 2025

    arXiv 2025

  59. [59]

    H. Li, S. Yang, Y . Chen, Y . Tian, X. Yang, X. Chen, H. Wang, T. Wang, F. Zhao, D. Lin, et al. Cronusvla: Transferring latent motion across time for multi-frame prediction in manipulation. arXiv e-prints, pages arXiv–2506, 2025

    2025

  60. [60]

    Y . Gao, J. Liu, S. Li, and S. Song. Gated memory policy.arXiv preprint arXiv:2604.18933, 2026

    Pith/arXiv arXiv 2026

  61. [61]

    H. Lei, W. Song, H. Zhang, J. Pei, J. Chen, H. Yan, H. Zhao, P. Ding, Z. Zhang, L. Huang, et al. Robomemarena: A comprehensive and challenging robotic memory benchmark.arXiv preprint arXiv:2605.10921, 2026

    Pith/arXiv arXiv 2026

  62. [62]

    Black, N

    K. Black, N. Brown, J. Darpinian, K. Dhabalia, D. Driess, A. Esmail, M. Equi, C. Finn, N. Fusai, et al. π0.5: A vision-language-action model with open-world generalization.arXiv preprint arXiv:2504.16054, 2025

    Pith/arXiv arXiv 2025

  63. [63]

    T. Wan, A. Wang, B. Ai, B. Wen, C. Mao, C.-W. Xie, D. Chen, F. Yu, H. Zhao, J. Yang, J. Zeng, J. Wang, J. Zhang, J. Zhou, J. Wang, J. Chen, K. Zhu, K. Zhao, K. Yan, L. Huang, M. Feng, N. Zhang, P. Li, P. Wu, R. Chu, R. Feng, S. Zhang, S. Sun, T. Fang, T. Wang, T. Gui, T. Weng, T. Shen, W. Lin, W. Wang, W. Wang, W. Zhou, W. Wang, W. Shen, W. Yu, X. Shi, X....

    Pith/arXiv arXiv 2025

  64. [64]

    Liang, L

    W. Liang, L. Yu, L. Luo, S. Iyer, N. Dong, C. Zhou, G. Ghosh, M. Lewis, W.-t. Yih, L. Zettle- moyer, et al. Mixture-of-transformers: A sparse and scalable architecture for multi-modal foundation models.arXiv preprint arXiv:2411.04996, 2024

    Pith/arXiv arXiv 2024

  65. [65]

    Huang, Z

    X. Huang, Z. Li, G. He, M. Zhou, and E. Shechtman. Self forcing: Bridging the train-test gap in autoregressive video diffusion. InAdvances in Neural Information Processing Systems, 2025

    2025

  66. [66]

    L. X. Shi, B. Ichter, M. Equi, L. Ke, K. Pertsch, Q. Vuong, J. Tanner, A. Walling, H. Wang, N. Fusai, et al. Hi robot: Open-ended instruction following with hierarchical vision-language- action models.arXiv preprint arXiv:2502.19417, 2025

    Pith/arXiv arXiv 2025

  67. [67]

    A. Figure. Helix: A vision-language-action model for generalist humanoid control.Figure AI News, 2024

    2024

  68. [68]

    C. Deng, D. Zhu, K. Li, C. Gou, F. Li, Z. Wang, S. Zhong, W. Yu, X. Nie, Z. Song, G. Shi, and H. Fan. Emerging properties in unified multimodal pretraining.arXiv preprint arXiv:2505.14683, 2025. 13 A Appendix A.1 Implementation Details Training setup.Training is conducted in bfloat16 mixed precision with FSDP, activation checkpoint- ing on every DiT block...

    Pith/arXiv arXiv 2025