Pith. sign in

REVIEW 2 major objections 1 minor 34 references

BehaviorVLA builds unified long-horizon behavior representations from trajectories using a causal Mamba encoder and decodes them via phase alignment to improve VLA robustness under shifts.

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

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

T0 review · grok-4.3

2026-06-30 16:49 UTC pith:UJUUTGZF

load-bearing objection BehaviorVLA combines causal Mamba for long-horizon encoding with phase-conditioned decoding to target generalization and data efficiency in VLA models, and reports solid benchmark numbers plus a 50% data sim-to-real match. the 2 major comments →

arxiv 2605.22671 v2 pith:UJUUTGZF submitted 2026-05-21 cs.CV

From Abstraction to Instantiation: Learning Behavioral Representation for Vision-Language-Action Model

Bing Hu , Zaijing Li , Rui Shao , Junda Chen , April Hua Liu , Wei-Shi Zheng , Liqiang Nie This is my paper
classification cs.CV
keywords vision-language-actionbehavioral representationMambamanipulationsim-to-real transfertemporal coherencephase conditioningrobotic control
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.

The paper claims that current vision-language-action models produce inconsistent actions in new environments because their latent representations break tasks into short fragments and fail to adjust to execution progress. It introduces BehaviorVLA as a fix, built around two matching parts: a visuomotor encoder that folds entire trajectories into one coherent representation and a decoder that reads that representation while tracking how far the task has advanced. The resulting model reaches higher success rates on three manipulation benchmarks and matches a strong baseline in real-world transfer while using half the demonstration data.

Core claim

BehaviorVLA learns temporally coherent behavioral representations by pairing a Visuomotor Behavior Encoder that uses causal Mamba layers to compress long-horizon trajectory information into a single vector with a Phase-conditioned Behavior Decoder that produces actions by continuously matching task-level priors to the current stage of execution. This design directly targets short-horizon fragmentation and static alignment problems in prior action-centric latent approaches.

What carries the argument

The paired Visuomotor Behavior Encoder (VBE), a causal Mamba network that aggregates full trajectories, and Phase-conditioned Behavior Decoder (PBD), which conditions action output on execution phase to align priors with progress.

Load-bearing premise

The main barriers to robust VLA behavior are short-horizon fragmentation and static alignment, and replacing them with long-horizon causal aggregation plus dynamic phase decoding will produce consistent actions when environments change.

What would settle it

Run BehaviorVLA on a new set of manipulation tasks that deliberately vary trajectory length or introduce distribution shifts outside the RoboTwin 2.0, LIBERO, and CALVIN suites and check whether success rates fall below those of OpenVLA-OFT or similar baselines.

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

If this is right

  • The model achieves 58% success on RoboTwin 2.0, 98% on LIBERO, and 4.36 average length on CALVIN.
  • In sim-to-real transfer the method reaches the same performance as OpenVLA-OFT while trained on only 50% of the demonstration data.
  • Temporally coherent representations reduce performance degradation when test conditions differ from training.
  • The symmetric encoder-decoder structure supports both representation learning and real-time action generation within one framework.

Where Pith is reading between the lines

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

  • If the phase-conditioned decoder generalizes, similar alignment mechanisms could improve consistency in other long-sequence prediction settings such as video generation or autonomous driving.
  • The data-efficiency result suggests that explicit long-horizon encoding may lower the volume of robot demonstrations needed for new tasks.
  • Replacing transformers with causal Mamba in the encoder may enable faster inference on edge hardware for real-time control loops.

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

2 major / 1 minor

Summary. The manuscript proposes BehaviorVLA, a Vision-Language-Action framework consisting of a Visuomotor Behavior Encoder (VBE) that uses a causal Mamba architecture to aggregate long-horizon trajectory information into a unified behavior representation, and a Phase-conditioned Behavior Decoder (PBD) that decodes this representation into actions by dynamically aligning task-level priors with execution progress. It claims these components address short-horizon temporal fragmentation and static execution-alignment in prior action-centric latent variable approaches, yielding state-of-the-art success rates of 58% on RoboTwin 2.0, 98% on LIBERO, and 4.36 average length on CALVIN, plus matching OpenVLA-OFT performance in sim-to-real transfer using only 50% of the demonstration data.

Significance. If the empirical claims hold with proper controls, the work could advance VLA models by demonstrating that long-horizon causal aggregation and phase-conditioned decoding improve robustness and data efficiency under distribution shifts in robotic manipulation tasks. The symmetric encoder-decoder design and use of Mamba for temporal coherence represent a targeted architectural response to identified limitations in existing methods.

major comments (2)
  1. [Abstract] Abstract: The central claims of state-of-the-art performance (58%, 98%, 4.36 Avg.Len) and data-efficient sim-to-real transfer are presented without any reference to the specific baseline methods, number of evaluation trials, error bars, statistical tests, or train/test splits. These details are load-bearing for assessing whether the reported numbers support superiority over prior VLA approaches.
  2. [Introduction / Method description] The motivating premise that short-horizon temporal fragmentation and static execution-alignment are the primary limitations of existing approaches (and that causal Mamba aggregation plus phase-conditioning will resolve inconsistent behaviors under shifts) is stated but lacks quantitative justification via targeted ablations or controlled distribution-shift experiments that isolate these factors from other variables such as model scale or training data volume.
minor comments (1)
  1. [Abstract] The abstract uses 'temporally coherent behavioral representations' (plural) in one sentence and 'a unified behavior representation' (singular) in the next; consistent terminology would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major comment below and indicate planned revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claims of state-of-the-art performance (58%, 98%, 4.36 Avg.Len) and data-efficient sim-to-real transfer are presented without any reference to the specific baseline methods, number of evaluation trials, error bars, statistical tests, or train/test splits. These details are load-bearing for assessing whether the reported numbers support superiority over prior VLA approaches.

    Authors: We agree that the abstract would benefit from additional context. In the revised manuscript we will update the abstract to reference the primary baselines (e.g., OpenVLA-OFT, RT-2) and note that full details on evaluation trials, error bars, statistical tests, and train/test splits appear in Section 4 and the supplementary material. revision: yes

  2. Referee: [Introduction / Method description] The motivating premise that short-horizon temporal fragmentation and static execution-alignment are the primary limitations of existing approaches (and that causal Mamba aggregation plus phase-conditioning will resolve inconsistent behaviors under shifts) is stated but lacks quantitative justification via targeted ablations or controlled distribution-shift experiments that isolate these factors from other variables such as model scale or training data volume.

    Authors: The design rationale is grounded in observed failure modes of prior VLA models. While overall gains are shown across benchmarks, the manuscript does not contain targeted ablations that isolate temporal fragmentation and static alignment under controlled distribution shifts while holding model scale and data volume fixed. We will add these ablation studies in the revision. revision: yes

Circularity Check

0 steps flagged

No significant circularity; claims rest on external benchmark results

full rationale

The abstract and available text present BehaviorVLA as a proposed architecture (VBE with causal Mamba, PBD with phase conditioning) whose value is demonstrated via success rates on named external datasets (RoboTwin 2.0, LIBERO, CALVIN) and a sim-to-real comparison against OpenVLA-OFT. No equations, fitted parameters, or self-citations are shown that would reduce any claimed prediction or uniqueness result to the inputs by construction. Performance numbers are reported as measured outcomes rather than quantities derived from internal fits, satisfying the self-contained criterion.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 2 invented entities

Review is based solely on the abstract; full details on any fitted hyperparameters, background assumptions, or additional components are unavailable. The ledger therefore records only the high-level assumptions and newly named architectural elements stated in the abstract.

axioms (2)
  • domain assumption Causal Mamba architecture can aggregate long-horizon trajectory information into a unified behavior representation that improves generalization
    Invoked in the description of the Visuomotor Behavior Encoder component.
  • domain assumption Dynamically aligning task-level priors with real-time execution progress via phase conditioning produces more precise actions
    Invoked in the description of the Phase-conditioned Behavior Decoder component.
invented entities (2)
  • Visuomotor Behavior Encoder (VBE) no independent evidence
    purpose: Aggregate long-horizon trajectory information into a unified behavior representation using causal Mamba
    Newly proposed architectural component in the BehaviorVLA framework.
  • Phase-conditioned Behavior Decoder (PBD) no independent evidence
    purpose: Decode the behavior representation into precise actions by dynamic alignment of task priors and execution progress
    Newly proposed architectural component in the BehaviorVLA framework.

pith-pipeline@v0.9.1-grok · 5780 in / 1751 out tokens · 67585 ms · 2026-06-30T16:49:36.745828+00:00 · methodology

0 comments
read the original abstract

Vision-Language-Action (VLA) models often suffer from performance degradation under distribution shifts, as they struggle to learn generalized behavior representations across varying environments. While existing approaches attempt to construct behavior representations through action-centric latent variables, they are often limited by short-horizon temporal fragmentation and static execution-alignment, leading to inconsistent behaviors in complex scenarios. To address these limitations, we propose \textbf{BehaviorVLA}, a framework that facilitates robust manipulation through the learning of a temporally coherent behavioral representations. Our approach features two symmetric components: (1) the \textbf{Visuomotor Behavior Encoder (VBE)}, which utilizes a causal Mamba-based architecture to aggregate long-horizon trajectory information into a unified behavior representation; and (2) the \textbf{Phase-conditioned Behavior Decoder (PBD)}, which decodes this representation into precise actions by dynamically aligning task-level priors with real-time execution progress. Experiments on RoboTwin 2.0, LIBERO, and CALVIN demonstrate state-of-the-art success rates of 58\%, 98\%, and 4.36 (Avg.Len), respectively. Notably, in real-world sim-to-real transfer, BehaviorVLA matches the performance of OpenVLA-OFT using only 50\% of the demonstration data, showcasing its superior data efficiency and generalization.

Figures

Figures reproduced from arXiv: 2605.22671 by April Hua Liu, Bing Hu, Junda Chen, Liqiang Nie, Rui Shao, Wei-Shi Zheng, Zaijing Li.

Figure 1
Figure 1. Figure 1: (a) Motivation: Standard VLAs learn mappings in high-dimensional space without explicit manifold constraints. In contrast, our goal is to learn a low-dimensional behavioral man￾ifold to capture transferable patterns. (b) Architecture: Unlike standard VLAs, BehaviorVLA incorporates the Visuomotor Behav￾ior Encoder (VBE), Phase-conditioned Behavior Decoder (PBD), and Behavior Memory Bank to learn and retriev… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of BehaviorVLA. Given an instruction and observation, the Vision-Language backbone first integrates multimodal information to retrieve a global prototype zproto from the Memory Bank. The retrieval is performed only once at the beginning of each episode, and the retrieved prototype remains fixed during execution as a stable behavioral prior. Simultaneously, the Visuomotor Behavior Encoder models th… view at source ↗
Figure 3
Figure 3. Figure 3: Real-world task setup and evaluation results. BehaviorVLA outperforms OpenVLA-OFT(Kim et al., 2025) and π0.5 (Intelligence et al., 2025) across both generalization and long-horizon tasks. Notably, BehaviorVLA demonstrates superior data efficiency, maintaining competitive performance even when trained with reduced dataset sizes (50% and 75%). achieves an average success rate of 98%, outperforming existing s… view at source ↗
Figure 4
Figure 4. Figure 4: Ablation on Guidance Strength λ in the inference. An optimal guidance strength is essential. Either insufficient or exces￾sive λ leads to degradation. 34%. This gain highlights the critical role of the Phase￾conditioned Behavior Decoder (PBD). By continuously aligning the action generation with the real-time execution phase, PBD prevents the temporal drift often observed in standard policies, ensuring cons… view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison of simulation and real-world manipulation tasks. Yellow bounding boxes indicate the training scenarios. Top: In simulation, the baseline π0.5 (Black et al., 2025) fails to grasp the target block when subjected to variations in background and object position. Bottom: In real-world experiments, our BehaviorVLA demonstrates strong few-shot transfer capabilities, accurately completing th… view at source ↗
Figure 6
Figure 6. Figure 6: t-SNE Visualization. (a) The VBE shows clear, distinct behavior clusters. Removing (b) the vision stream or (c) the action stream causes clusters to mix and scatter. This highlights our tri-stream design is essential for learning highly discriminative behavior representations. trajectory generation. Conversely, an excessively large λ imposes an over-constraining prior that suppresses the fine￾grained local… view at source ↗
Figure 7
Figure 7. Figure 7: t-SNE Visualization. (a) The VBE shows clear, distinct behavior clusters. Removing (b) the vision stream or (c) the action stream causes clusters to mix and scatter. This highlights our tri-stream design is essential for learning highly discriminative behavior representations. Results on Real-World. As illustrated in [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative results of BehaviorVLA on Real-World. From top to bottom, we illustrate four Generalization Tasks: Adjust bottle, Stack bowl on plate, Place bread in basket, and Place basket on tablecloth.The model demonstrates robust adaptability in scenarios requiring precise interaction, confirming the effectiveness of the learned visuomotor behavior manifold. Move the blocks to the center of the table, and… view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative results of BehaviorVLA on Real-World. From top to bottom, we illustrate four Long-horizon Tasks: Move and stack blocks on center, Place containers on plate, Pick and place blocks in bowl, and Place bottles and cans in basket.By conditioning the policy on a global prototype for structural guidance and dynamically tracking execution via phase variables, BehaviorVLA mitigates temporal drift, ensur… view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative results of BehaviorVLA on Real-World. From top to bottom, we illustrate four Long-horizon Tasks: Move and stack blocks on center, Place containers on plate, Pick and place blocks in bowl, and Place bottles and cans in basket.By conditioning the policy on a global prototype for structural guidance and dynamically tracking execution via phase variables, BehaviorVLA mitigates temporal drift, ensur… view at source ↗

Review history (2 revisions) →

discussion (0)

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

Reference graph

Works this paper leans on

34 extracted references · 29 canonical work pages · 16 internal anchors

  1. [1]

    Dexterous manipulation through imitation learning: A survey.arXiv preprint arXiv:2504.03515, 2025

    9 From Abstraction to Instantiation: Learning Behavioral Representation for Vision-Language-Action Model An, S., Meng, Z., Tang, C., Zhou, Y ., Liu, T., Ding, F., Zhang, S., Mu, Y ., Song, R., Zhang, W., et al. Dexterous manipulation through imitation learning: A survey.arXiv preprint arXiv:2504.03515,

    work page arXiv

  2. [2]

    GR00T N1: An Open Foundation Model for Generalist Humanoid Robots

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

    work page internal anchor Pith review Pith/arXiv arXiv

  3. [4]

    $\pi_0$: A Vision-Language-Action Flow Model for General Robot Control

    doi: 10.48550. arXiv preprint ARXIV .2410.24164. Black, K., Brown, N., Driess, D., Esmail, A., Equi, M., Finn, C., Fusai, N., Groom, L., Hausman, K., Ichter, B., et al. pi 0: A vision-language-action flow model for general robot control.arXiv preprint arXiv:2410.24164,

    work page internal anchor Pith review Pith/arXiv arXiv

  4. [5]

    arXiv preprint arXiv:2410.08001 (2024) 4, 6

    Bu, Q., Li, H., Chen, L., Cai, J., Zeng, J., Cui, H., Yao, M., and Qiao, Y . Towards synergistic, generalized, and efficient dual-system for robotic manipulation.arXiv preprint arXiv:2410.08001,

    work page arXiv

  5. [6]

    UniVLA: Learning to Act Anywhere with Task-centric Latent Actions

    Bu, Q., Yang, Y ., Cai, J., Gao, S., Ren, G., Yao, M., Luo, P., and Li, H. Univla: Learning to act anywhere with task- centric latent actions.arXiv preprint arXiv:2505.06111,

    work page internal anchor Pith review Pith/arXiv arXiv

  6. [7]

    arXiv preprint arXiv:2506.01953 (2025) 2

    Chen, G., Zhou, X., Shao, R., Lyu, Y ., Zhou, K., Wang, S., Li, W., Li, Y ., Qi, Z., and Nie, L. Less is more: Em- powering gui agent with context-aware simplification. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 5901–5911, 2025a. Chen, H., Liu, J., Gu, C., Liu, Z., Zhang, R., Li, X., He, X., Guo, Y ., Fu, C.-W., Zhang,...

    work page arXiv

  7. [8]

    Video Prediction Policy: A Generalist Robot Policy with Predictive Visual Representations

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

    work page internal anchor Pith review Pith/arXiv arXiv

  8. [9]

    $\pi_{0.5}$: a Vision-Language-Action Model with Open-World Generalization

    Intelligence, P., Black, K., Brown, N., Darpinian, J., Dha- balia, K., Driess, D., Esmail, A., Equi, M., Finn, C., Fusai, N., et al. pi05: a vision-language-action model with open- world generalization.arXiv preprint arXiv:2504.16054,

    work page internal anchor Pith review Pith/arXiv arXiv

  9. [10]

    OpenVLA: An Open-Source Vision-Language-Action Model

    Kim, M. J., Pertsch, K., Karamcheti, S., Xiao, T., Balakr- ishna, A., Nair, S., Rafailov, R., Foster, E., Lam, G., San- keti, P., et al. Openvla: An open-source vision-language- action model.arXiv preprint arXiv:2406.09246,

    work page internal anchor Pith review Pith/arXiv arXiv

  10. [11]

    Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success

    Kim, M. J., Finn, C., and Liang, P. Fine-tuning vision- language-action models: Optimizing speed and success. arXiv preprint arXiv:2502.19645,

    work page internal anchor Pith review Pith/arXiv arXiv

  11. [12]

    Behavior generation with latent actions,

    Lee, S., Wang, Y ., Etukuru, H., Kim, H. J., Shafiullah, N. M. M., and Pinto, L. Behavior generation with latent actions.arXiv preprint arXiv:2403.03181,

    work page arXiv

  12. [13]

    Pointvla: Injecting the 3d world into vision-language- action models.arXiv preprint arXiv:2503.07511, 2025a

    Li, C., Wen, J., Peng, Y ., Peng, Y ., Feng, F., and Zhu, Y . Pointvla: Injecting the 3d world into vision-language- action models.arXiv preprint arXiv:2503.07511, 2025a. Li, H., Lv, Q., Shao, R., Deng, X., Li, Y ., Hao, J., and Nie, L. Star: Learning diverse robot skill abstractions through rotation-augmented vector quantization.arXiv preprint arXiv:2506...

    work page arXiv 2022

  13. [14]

    Optimus-3: Towards generalist multi- modal minecraft agents with scalable task experts.arXiv preprint arXiv:2506.10357, 2025f

    Li, Z., Xie, Y ., Shao, R., Chen, G., Guan, W., Jiang, D., and Nie, L. Optimus-3: Towards generalist multi- modal minecraft agents with scalable task experts.arXiv preprint arXiv:2506.10357, 2025f. Li, Z., Xie, Y ., Shao, R., Chen, G., Jiang, D., and Nie, L. Optimus-2: Multimodal minecraft agent with goal- observation-action conditioned policy. InProceedi...

    work page arXiv

  14. [15]

    Echovla: Robotic vision-language-action model with synergistic declarative memory for mobile manipulation,

    URL https: //arxiv.org/abs/2511.18112. Lin, T., Zhang, Y ., Li, Q., Qi, H., Yi, B., Levine, S., and Malik, J. Learning visuotactile skills with two multifin- gered hands. In2025 IEEE International Conference on Robotics and Automation (ICRA), pp. 5637–5643. IEEE,

    work page arXiv

  15. [16]

    Liu, H., Li, C., Wu, Q., and Lee, Y . J. Visual instruction tun- ing.Advances in neural information processing systems, 36, 2024a. Liu, H., Li, X., Li, P., Liu, M., Wang, D., Liu, J., Kang, B., Ma, X., Kong, T., and Zhang, H. Towards generalist robot policies: What matters in building vision-language-action models. 2025a. Liu, J., Liu, M., Wang, Z., An, P...

    work page internal anchor Pith review Pith/arXiv arXiv

  16. [17]

    FAST: Efficient Action Tokenization for Vision-Language-Action Models

    Pertsch, K., Stachowicz, K., Ichter, B., Driess, D., Nair, S., Vuong, Q., Mees, O., Finn, C., and Levine, S. Fast: Efficient action tokenization for vision-language-action models.arXiv preprint arXiv:2501.09747,

    work page internal anchor Pith review Pith/arXiv arXiv

  17. [18]

    SpatialVLA: Exploring Spatial Representations for Visual-Language-Action Model

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

    work page internal anchor Pith review Pith/arXiv arXiv

  18. [19]

    Large VLM-based Vision-Language-Action Models for Robotic Manipulation: A Survey

    Shao, R., Li, W., Zhang, L., Zhang, R., Liu, Z., Chen, R., and Nie, L. Large vlm-based vision-language-action models for robotic manipulation: A survey.arXiv preprint arXiv:2508.13073,

    work page internal anchor Pith review Pith/arXiv arXiv

  19. [20]

    Hats: Hardness-aware trajectory syn- thesis for gui agents.arXiv preprint arXiv:2603.12138,

    Shao, R., Gao, R., Xie, B., Li, Y ., Zhou, K., Wang, S., Guan, W., and Chen, G. Hats: Hardness-aware trajectory syn- thesis for gui agents.arXiv preprint arXiv:2603.12138,

    work page arXiv

  20. [21]

    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., and Huang, G. Memo- ryvla: Perceptual-cognitive memory in vision-language- action models for robotic manipulation.arXiv preprint arXiv:2508.19236,

    work page internal anchor Pith review Pith/arXiv arXiv

  21. [22]

    SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics

    Shukor, M., Aubakirova, D., Capuano, F., Kooijmans, P., Palma, S., Zouitine, A., Aractingi, M., Pascal, C., Russi, M., Marafioti, A., et al. Smolvla: A vision-language- action model for affordable and efficient robotics.arXiv preprint arXiv:2506.01844,

    work page internal anchor Pith review Pith/arXiv arXiv

  22. [23]

    Reconvla: Reconstructive vision-language-action model as effective robot perceiver, 2025.https://arxiv.org/abs/2508.10333

    Song, W., Zhou, Z., Zhao, H., Chen, J., Ding, P., Yan, H., Huang, Y ., Tang, F., Wang, D., and Li, H. Reconvla: Reconstructive vision-language-action model as effective robot perceiver.arXiv preprint arXiv:2508.10333,

    work page arXiv

  23. [24]

    Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation

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

    work page internal anchor Pith review Pith/arXiv arXiv

  24. [25]

    Vq-vla: Improving vision-language-action models via scaling vector-quantized action tokenizers,

    Wang, Y ., Zhu, H., Liu, M., Yang, J., Fang, H.-S., and He, T. Vq-vla: Improving vision-language-action models via scaling vector-quantized action tokenizers.arXiv preprint arXiv:2507.01016,

    work page arXiv

  25. [26]

    Embodied-RAG: General non-parametric Embodied Memory for Retrieval and Generation

    Xie, Q., Min, S. Y ., Ji, P., Yang, Y ., Zhang, T., Xu, K., Ba- jaj, A., Salakhutdinov, R., Johnson-Roberson, M., and Bisk, Y . Embodied-rag: General non-parametric embod- ied memory for retrieval and generation.arXiv preprint arXiv:2409.18313,

    work page arXiv

  26. [27]

    13 Renjun Xu and Yang Yan

    Xie, Y ., Li, Z., Shao, R., Chen, G., Zhou, K., Li, Y ., Jiang, D., and Nie, L. Mirage-1: Augmenting and updating gui agent with hierarchical multimodal skills.arXiv preprint arXiv:2506.10387,

    work page arXiv

  27. [28]

    RoboSSM: Scalable In-context Imitation Learning via State-Space Models

    12 From Abstraction to Instantiation: Learning Behavioral Representation for Vision-Language-Action Model Yoo, Y ., Hu, J., Zhu, Y ., Liu, B., Liu, Q., Mart´ın-Mart´ın, R., and Stone, P. Robossm: Scalable in-context imi- tation learning via state-space models.arXiv preprint arXiv:2509.19658,

    work page internal anchor Pith review Pith/arXiv arXiv

  28. [29]

    3D Diffusion Policy: Generalizable Visuomotor Policy Learning via Simple 3D Representations

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

    work page internal anchor Pith review Pith/arXiv arXiv

  29. [30]

    Up-vla: A unified understanding and prediction model for embodied agent.arXiv preprint arXiv:2501.18867, 2025

    Zhang, J., Guo, Y ., Hu, Y ., Chen, X., Zhu, X., and Chen, J. Up-vla: A unified understanding and prediction model for embodied agent.arXiv preprint arXiv:2501.18867, 2025a. Zhang, R., Shao, R., Chen, G., Zhang, M., Zhou, K., Guan, W., and Nie, L. Falcon: Resolving visual redundancy and fragmentation in high-resolution multimodal large language models via...

    work page arXiv

  30. [31]

    13 From Abstraction to Instantiation: Learning Behavioral Representation for Vision-Language-Action Model A. Related Work Vision-Language-Action Models.Building upon the advancements of Vision-Language Models (VLMs) (Alayrac et al., 2022; Shao et al., 2019; Karamcheti et al., 2024; Shao et al., 2023; 2024; Li et al., 2025d; Zhou et al., 2025a; Zhang et al...

    2022

  31. [32]

    have emerged as a promising paradigm in robot learning. Recent works have further extended VLA capabilities through the integration of enhanced visual perception(Li et al., 2025a; Qu et al., 2025; Liu et al., 2025b), efficient paradigms (Liu et al., 2024b; Chen et al., 2025b), and dual-system architectures (Bjorck et al., 2025; Wang et al., 2025; Wen et a...

    2025

  32. [33]

    MAP-VLA(Li et al., 2025c) further reduces fragment inconsistency through stage-wise segmentation and alignment

    represents scene and episodic information as declarative memory for retrieval and fusion. MAP-VLA(Li et al., 2025c) further reduces fragment inconsistency through stage-wise segmentation and alignment. Related ideas also appear in embodied agents and generalist policies (Zhu et al., 2024; Anwar et al., 2025; Xie et al., 2024), which retrieve trajectories ...

    2024

  33. [34]

    Recent VLAs(Black et al., 2025; Black et al.; Shukor et al.,

    have become the standard for robot control, modeling generation as a transport process from gaussian noise to multi-modal distributions. Recent VLAs(Black et al., 2025; Black et al.; Shukor et al.,

    2025

  34. [35]

    distribution shift

    by 11%. In the challenging ABC → D setting (unseen environments), standard VLA models suffer from the “distribution shift” problem, overfitting to specific simulation textures. In contrast, our results validate the effectiveness of the Specific-to- General Abstraction. By distilling trajectories into a scene-invariant behavior manifold, VBE filters out en...

    2022