Pith. sign in

REVIEW 3 major objections 4 minor 43 references

TEMPO: Semantic-Action Decoupled RL Post-Training for Vision-Language-Action Models

T0 review · 3 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read TEMPO claims that decoupling the update frequencies of a VLA model's semantic projection layer and action expert—slow for semantics, fast for control—improves long-horizon manipulation, reaching 81.7% success on CALVIN ABC→D.

desk verdict A genuinely new design idea for VLA RL post-training, with plausible ablations but underwhelming statistical evidence and an unaddressed non-stationarity in the semantic critic. read the letter →

arxiv 2608.07314 v1 pith:WTIZHG2P submitted 2026-08-07 cs.RO cs.CV

Ziheng Liu , Quantao Yang This is my paper · ORCID
classification cs.ROcs.CV
keywords vision-language-actionmodelsreinforcementlearningpost-trainingrobotmanipulationTD3semantic-actiondecouplingtwo-timescaleupdateslong-horizonCALVINbenchmark
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

TEMPO is a post-training recipe for vision-language-action (VLA) robot policies that treats semantic understanding and low-level control as separate learning problems with separate learning rates. The paper claims that freezing the pretrained vision-language backbone, adapting the semantic projection layer in a slow reinforcement-learning loop, and adapting the action expert in a fast loop gives better long-horizon manipulation than updating both modules together. This matters because the standard alternatives—supervised fine-tuning or uniform RL post-training—either suffer from distribution mismatch or destabilize the latent action space that the action head learns on. On CALVIN ABC→D the method reaches 81.7% five-task success versus 78.4% for the single-loop RL baseline, and it reaches and holds higher evaluation rewards on two real-world multi-stage tasks.

What carries the argument

The central mechanism is two module-level TD3 loops sharing a single rollout stream but no gradients. One loop treats the multimodal semantic representation $h_t=\mathrm{VL}(o_t,l_t)$ as its state and the latent action $z_t=\pi_\theta^s(h_t)$ as its action; the other treats $z_t$ as its state and the robot action chunk $a_t=\pi_\phi^a(z_t)$ as the executed action. The semantic projection layer is updated every $d_s$ critic updates and the action expert every $d_a$ updates, so the relative frequency is $\rho=f_a/f_s=N_a d_s/(N_s d_a)$; the paper finds $\rho>1$, especially $5:1$, is what converts RL updates into long-horizon gains. Stop-gradients and separate target networks keep the two gradient pathways independent, and the frozen vision-language backbone preserves the pretrained semantics that both loops consume.

What would settle it

Run TEMPO at $\rho=1:1$ and $\rho=5:1$ with matched total numbers of critic updates; if the equal-frequency condition converges to the same or higher success, then the frequency decoupling is not the causal mechanism behind the reported gain.

Watch

Extended reading notes

Core claim

TEMPO's central claim is that applying RL to a VLA policy through two module-level TD3 loops, one for the semantic projection layer $\pi_\theta^s$ and one for the action expert $\pi_\phi^a$, with an update-frequency ratio $\rho=f_a/f_s>1$, improves long-horizon manipulation over both the pretrained policy and a single-loop RL baseline. At $\rho=5:1$ the paper reports 100.0/97.1/92.9/87.1/81.7 percent success on CALVIN ABC→D tasks 1–5 with an average chain length of 4.59, against 99.4/96.7/91.3/85.5/78.4 for FLOWER-RL and 99.4/95.8/90.7/84.9/77.8 for the pretrained FLOWER reference. The proposed mechanism is latent-action stability: infrequent semantic updates keep the distribution of $z_t=\pi_\theta^s(h_t)$ comparatively stable, while frequent action-expert updates let control absorb online feedback; at the equal-frequency setting $\rho=1:1$ the gain over FLOWER disappears, which the paper reads as evidence that the frequency decoupling, not RL updates alone, carries the improvement.

Load-bearing premise

Everything rests on the assumption that the semantic-level TD3 loop's model of how latent states evolve stays valid while the action expert is being updated in the other loop.

Editorial extensions

If this is right

  • RL post-training can improve long-horizon VLA manipulation without backpropagating through the vision-language backbone, preserving pretrained semantic knowledge and cutting compute.
  • The action-to-semantic frequency ratio $\rho$ becomes a practical tuning knob: $\rho=1:1$ forfeits the gain, $5:1$ gives the best result, and $10:1$ remains competitive but slightly worse.
  • Because both module-level loops consume the same environment rollouts, decoupling update frequencies requires no extra data collection—only a change in how often each network is updated.
  • The gains grow as the instruction chain lengthens, from +0.6 points at SR1 to +3.9 at SR5, indicating the mechanism specifically offsets error accumulation over long horizons.

Reading between the lines

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

  • A testable extension of the claimed mechanism: log latent-action drift under $\rho=1:1$ versus $\rho=5:1$; if the slower semantic loop reduces drift, that directly supports the stability story.
  • The frequency-decoupling principle may transfer to any modular policy, not just FLOWER, and to asynchronous inference: updating semantically meaningful representations more slowly than low-level controllers is a general stability strategy.
  • A natural extension is to make $\rho$ adaptive during training, starting near equality and widening as the action expert improves, which could capture the reported benefits without manual grid search.
  • The single-task post-training setting outperforming the 34-task setting suggests the interaction budget is diluted across tasks; adaptive allocation of rollouts by task difficulty could be a further improvement.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. This paper proposes TEMPO, a two-timescale reinforcement learning (RL) post-training method for vision-language-action (VLA) models. The vision-language backbone is frozen, and two modules—a semantic projection layer and an action expert—are optimized with separate TD3 loops. The key design choice is to update the semantic projection layer less frequently than the action expert, with a reported action-to-semantic update-frequency ratio ρ = 5:1 or 10:1, to reduce latent-action drift while allowing the action expert to adapt rapidly to online feedback. The method is evaluated on the CALVIN ABC→D benchmark and two real-world manipulation tasks. The paper reports that TEMPO achieves higher success rates than the pretrained FLOWER policy and a single-loop RL baseline (FLOWER-RL), and presents component ablations, frequency-ratio comparisons, and a task-set-size analysis.

Significance. The idea of decoupling update frequencies across functionally distinct modules in VLA post-training is plausible and relevant to the growing RL-for-VLA literature. The paper provides a clear architectural decomposition, component ablations, and real-robot experiments, which are useful empirical contributions. However, the theoretical soundness of the semantic-level TD3 loop under a non-stationary transition is not established, and the empirical evidence is weakened by the lack of variance reporting and the small absolute gains in the main comparison. If the authors can address these points, the design principle could be a valuable addition to the field; as written, the central claims are not yet fully supported.

major comments (3)
  1. [Section III-C, Eq. (5)] The semantic-level TD3 loop is defined over transitions (h_t, z_t, r_t, h_{t+1}, d_t), but the transition distribution depends on the action chunk a_t = π^a_φ(z_t), and φ is updated concurrently on a faster timescale. Thus the semantic-level transition and reward functions are non-stationary during training. The Bellman target in Eq. (5) bootstraps against this shifting environment without any justification that TD3 remains valid under such coupling. This is load-bearing because the claimed stability benefit of decoupled frequencies relies on the semantic critic producing reliable value estimates. The authors should either provide a formal condition (e.g., bounded φ-drift per semantic update) or empirical diagnostics (e.g., measuring semantic-critic TD error or value estimates under varying φ) to support the method's theoretical grounding.
  2. [Tables I-IV] No variance, number of seeds, or statistical significance tests are reported for the CALVIN experiments. The headline improvement over DeFI is 0.5 percentage points on SR5 (81.7 vs 81.2), which is likely within noise. The improvement over FLOWER-RL is 3.3 points, but without error bars it is unclear whether this difference is significant. Table III shows that the 1:1 dual-loop setting is only 0.1 point below FLOWER, so the benefit of the frequency decoupling is not established. Please report mean±std over at least 3–5 seeds for all runs and provide a significance test (e.g., paired bootstrap) for the key comparisons.
  3. [Section IV-C, Fig. 3] The real-world results report only shaded regions denoting one standard deviation across three seeds and give no numerical values at the evaluation checkpoints. The description that rewards 'gradually improve and stabilize' is qualitative. To support the claim that TEMPO reaches and maintains higher rewards than FLOWER-RL, the paper should report the mean±std success values or rewards at specific training steps for both tasks, and ideally per-seed counts of successful trials.
minor comments (4)
  1. [Section IV-A] The baseline list in Section IV-A omits RoboFlamingo, which appears in Table I; please add a brief description or citation for it.
  2. [Section III-E] The actual values of N_a, N_s, d_a, and d_s are not specified, so the frequency ratio ρ in Eq. (14) is not reproducible. Please state the per-round update counts and actor delay hyperparameters used for Tables III and IV.
  3. [Equations (7) and (12)] The notation z_t^sg in Eq. (12) is introduced only in the text; please define it explicitly in the equation or its surrounding caption.
  4. [Abstract] The abstract claims TEMPO 'consistently outperforms' both pretrained state-of-the-art VLA models and the RL post-training baseline. Given the 0.5 percentage-point gain over DeFI and the lack of statistical analysis, this wording is too strong; consider softening to 'achieves higher success rates in our experiments.'

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: TEMPO's claims are empirical and benchmark-grounded; the update-frequency ratio is a tuned hyperparameter, not a fitted prediction.

full rationale

The paper's central claims are supported by direct experimental comparisons on CALVIN and real-robot tasks, not by a derivation that reduces to its own assumptions. The semantic-level and action-level TD3 loops use standard Bellman targets (Eqs. 5 and 10) defined over stored transitions, and the semantic projection layer and action expert are distinct parameter sets with no gradient coupling. The action-to-semantic update-frequency ratio rho is a hyperparameter reported as a sweep (Table III), not a quantity fitted to the data and then relabeled as a prediction; the improvement of rho>1 over rho=1 is an empirical finding. The only self-citation is reference [17] (Visa-Flow), which includes an author of the present paper, but it is cited merely as related work on VLA models and plays no load-bearing role in the derivation of TEMPO's claims. The skeptical concern that the semantic-level transition is non-stationary because it depends on the evolving action expert is a technical correctness risk, not a circularity: it does not make any equation equal to its inputs by construction, nor does it rename a fitted parameter as a prediction. The paper is not self-referential in any way that would force its conclusions.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim depends on the two-loop MDP decomposition, the frozen-backbone semantic preservation assumption, and the choice of update-frequency ratio. The paper provides external benchmarks but no code or error bars, so the ledger mostly records assumptions rather than fitted constants.

free parameters (3)
  • Action-to-semantic update-frequency ratio rho = f_a:f_s = 5:1 (selected), also 10:1 and 1:1 evaluated
    The frequency ratio is the core contribution, but the optimal value is chosen from Table III on CALVIN without a theoretical derivation.
  • Per-round critic update counts and actor delays (N_a, N_s, d_a, d_s)
    Effective frequencies f_a and f_s are defined in Eq. (13), but concrete numeric values are never reported, making the actual update schedule non-reproducible.
  • RL post-training task-set size = 1 (best in Table IV, likely final)
    The task-set comparison shows single-task post-training performs best; the selected composition is an empirical choice.
assumptions (4)
  • domain assumption The semantic-level process is a stationary MDP with state h_t and action z_t.
    Section III-A and III-C; the transition h_t -> h_{t+1} actually depends on the concurrently updated action expert, so stationarity is not guaranteed.
  • domain assumption Freezing the pretrained vision-language backbone preserves the semantic knowledge needed for downstream tasks.
    Section III-B; based on prior work [12], [13] but still an assumption for this pipeline.
  • domain assumption Sparse binary task-completion reward is sufficient for both module-level loops.
    Section III-B and Section IV-C; no dense reward or auxiliary signal is used.
  • standard math TD3 is a valid base RL algorithm for both modules.
    Section III-C and III-D use TD3 from [14] without modification.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TEMPO: Semantic-Action Decoupled RL Post-Training for Vision-Language-Action Models." pith.science (2026). https://pith.science/paper/WTIZHG2P

@misc{pith2026260807314,
  author       = {Pith},
  title        = {Pith review of: TEMPO: Semantic-Action Decoupled RL Post-Training for Vision-Language-Action Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WTIZHG2P}},
  note         = {Machine review of arXiv:2608.07314}
}
read the original abstract

Vision-language-action (VLA) models are commonly adapted to downstream manipulation tasks via supervised fine-tuning (SFT) or online reinforcement learning (RL) post-training. SFT is prone to distribution mismatch, and existing RL approaches typically apply a single, uniform update strategy to all model components, ignoring their distinct functional roles. We propose TEMPO, a semantic-action decoupled, two-timescale RL post-training framework for VLA models. TEMPO freezes the pretrained vision-language backbone to preserve general semantic representations, and restricts adaptation to two components with dedicated RL optimization loops: the semantic projection layer and the low-level action expert. We update them at different rates--the semantic projection layer infrequently, to keep the latent action stable, and the action expert frequently, to rapidly incorporate control feedback from online interaction. This decoupling RL fine-tuning strategy prevents fast policy updates from destabilizing high-level semantic representations while still allowing the action expert to learn efficiently from online feedback. Experiments on the CALVIN benchmark and real-world manipulation tasks demonstrate that TEMPO consistently outperforms both pretrained state-of-the-art VLA models and the RL post-training baseline, while reaching and maintaining higher evaluation rewards on two real-world tasks.

Figures

Figures reproduced from arXiv: 2608.07314 by the authors.

Figure 1
Figure 1. Method overview. TEMPO keeps the VLM frozen and decouples the reinforcement learning update frequencies of the semantic projection layer and action expert, updating the former less frequently and the latter more frequently. limitations motivate online post-training paradigms that en￾able policies to adapt through task-specific feedback acquired from their own interactions with the environment. Online reinforcement l… view at source ↗
Figure 2
Figure 2. Framework of TEMPO. The frozen VLM encodes the instruction and observation into ht, which is transformed by the semantic projection layer into the latent action zt and subsequently mapped by the action expert to the robot action chunk at. Two separate module-level TD3 loops, each equipped with a dedicated replay buffer and critic, optimize the two modules using shared environment rollouts, with a lower update freque… view at source ↗
Figure 3
Figure 3. Real-world results on two manipulation tasks. (a) and (b) show the block-sliding task, while (c) and (d) show the car-retrieval task. (a) and (c) report the evaluation reward during online RL post-training, with shaded regions denoting one standard deviation across three independent random seeds. (b) and (d) present representative execution sequences. For block sliding, FLOWER-RL directly manipulates the block witho… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 26 canonical work pages

  1. [1]

    PaLM-E: An embodied multimodal language model,

    D. Driess, F. Xia, M. S. M. Sajjadi, C. Lynch, A. Chowdhery, B. Ichter, A. Wahid, J. Tompson, Q. Vuong, T. Yu, W. Huang, Y . Chebotar, P. Sermanet, D. Duckworth, S. Levine, V . Vanhoucke, K. Hausman, M. Toussaint, K. Greff, A. Zeng, I. Mordatch, and P. Florence, “PaLM-E: An embodied multimodal language model,” inProceedings of the 40th International Confe...

    work page 2023

  2. [2]

    RT-1: Robotics transformer for real-world control at scale,

    A. Brohan, N. Brown, J. Carbajal, Y . Chebotar, J. Dabis, C. Finn, K. Gopalakrishnan, K. Hausman, A. Herzog, J. Hsu, J. Ibarz, B. Ichter, A. Irpan, T. Jackson, S. Jesmonth, N. Joshi, R. Julian, D. Kalashnikov, Y . Kuang, I. Leal, K.-H. Lee, S. Levine, Y . Lu, U. Malla, D. Manjunath, I. Mordatch, O. Nachum, C. Parada, J. Peralta, E. Perez, K. Pertsch, J. Q...

    work page 2023

  3. [3]

    RT-2: Vision-language-action models transfer web knowledge to robotic control,

    B. Zitkovich, T. Yu, S. Xu, P. Xu, T. Xiao, F. Xia, J. Wu, P. Wohlhart, S. Welker, A. Wahid, Q. Vuong, V . Vanhoucke, H. Tran, R. Soricut, A. Singh, J. Singh, P. Sermanet, P. R. Sanketi, G. Salazar, M. S. Ryoo, K. Reymann, K. Rao, K. Pertsch, I. Mordatch, H. Michalewski, Y . Lu, S. Levine, L. Lee, T.-W. E. Lee, I. Leal, Y . Kuang, D. Kalashnikov, R. Julia...

    work page 2023

  4. [4]

    Open X-Embodiment: Robotic learning datasets and RT-X models,

    Open X-Embodiment Collaboration, “Open X-Embodiment: Robotic learning datasets and RT-X models,” in2024 IEEE International Conference on Robotics and Automation (ICRA), 2024, pp. 6892–6903. [Online]. Available: https://ieeexplore.ieee.org/document/ 10611477

    work page 2024

  5. [5]

    Octo: An open-source generalist robot policy,

    D. Ghosh, H. R. Walke, K. Pertsch, K. Black, O. Mees, S. Dasari, J. Hejna, T. Kreiman, C. Xu, J. Luo, Y . L. Tan, L. Y . Chen, Q. Vuong, T. Xiao, P. R. Sanketi, D. Sadigh, C. Finn, and S. Levine, “Octo: An open-source generalist robot policy,” inProceedings of Robotics: Science and Systems, Delft, Netherlands, July 2024. [Online]. Available: https://www.r...

    work page 2024

  6. [6]

    OpenVLA: An open-source vision- language-action model,

    M. J. Kim, K. Pertsch, S. Karamcheti, T. Xiao, A. Balakrishna, S. Nair, R. Rafailov, E. P. Foster, P. R. Sanketi, Q. Vuong, T. Kollar, B. Burchfiel, R. Tedrake, D. Sadigh, S. Levine, P. Liang, and C. Finn, “OpenVLA: An open-source vision- language-action model,” inProceedings of The 8th Conference on Robot Learning, ser. Proceedings of Machine Learning Re...

    work page 2025

  7. [7]

    Fine-tuning vision-language-action models: Optimizing speed and success,

    M. J. Kim, C. Finn, and P. Liang, “Fine-tuning vision-language-action models: Optimizing speed and success,” inProceedings of Robotics: Science and Systems, Los Angeles, CA, USA, June 2025. [Online]. Available: https://www.roboticsproceedings.org/rss21/p017.html

    work page 2025

  8. [8]

    A reduction of imitation learning and structured prediction to no-regret online learning,

    S. Ross, G. Gordon, and D. Bagnell, “A reduction of imitation learning and structured prediction to no-regret online learning,” in Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, ser. Proceedings of Machine Learning Research, vol. 15. PMLR, 2011, pp. 627–635. [Online]. Available: https://proceedings.mlr.pr...

    work page 2011

Show all 43 references
  1. [9]

    VLA-RL: Towards masterful and general robotic manipulation with scalable reinforcement learning,

    G. Lu, W. Guo, C. Zhang, Y . Zhou, H. Jiang, Z. Gao, Y . Tang, and Z. Wang, “VLA-RL: Towards masterful and general robotic manipulation with scalable reinforcement learning,” 2025. [Online]. Available: https://arxiv.org/abs/2505.18719

    2025 arXiv

  2. [10]

    SimpleVLA-RL: Scaling VLA training via reinforcement learning,

    H. Li, Y . Zuo, J. Yu, Y . Zhang, Z. Yang, K. Zhang, X. Zhu, Y . Zhang, T. Chen, G. Cui, D. Wang, D. Luo, Y . Fan, Y . Sun, J. Zeng, J. Pang, S. Zhang, Y . Wang, Y . Mu, B. Zhou, and N. Ding, “SimpleVLA-RL: Scaling VLA training via reinforcement learning,” inThe Fourteenth Int...

    2026

  3. [11]

    Rl token: Bootstrapping online rl with vision-language- action models,

    C. Xu, J. T. Springenberg, M. Equi, A. Amin, A. Esmail, S. Levine, and L. Ke, “Rl token: Bootstrapping online rl with vision-language- action models,”arXiv preprint arXiv:2604.23073, 2026

    2026 arXiv

  4. [12]

    Knowledge insulating vision-language-action models: Train fast, run fast, generalize better,

    D. Driess, J. T. Springenberg, B. Ichter, L. Yu, A. Li-Bell, K. Pertsch, A. Z. Ren, H. Walke, Q. Vuong, L. X. Shi, and S. Levine, “Knowledge insulating vision-language-action models: Train fast, run fast, generalize better,” inAdvances in Neural Information Processing Systems,...

    2025

  5. [13]

    Improving vision-language-action model with online reinforcement learning,

    Y . Guo, J. Zhang, X. Chen, X. Ji, Y .-J. Wang, Y . Hu, and J. Chen, “Improving vision-language-action model with online reinforcement learning,” 2025, accepted to ICRA 2025. [Online]. Available: https://arxiv.org/abs/2501.16664

    2025 arXiv

  6. [14]

    Addressing function approximation error in actor-critic methods,

    S. Fujimoto, H. van Hoof, and D. Meger, “Addressing function approximation error in actor-critic methods,” inProceedings of the 35th International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, vol. 80. PMLR, 2018, pp. 1587–1596. [Online]. Avail...

    2018

  7. [15]

    BridgeData V2: A dataset for robot learning at scale,

    H. R. Walke, K. Black, T. Z. Zhao, Q. Vuong, C. Zheng, P. Hansen-Estruch, A. W. He, V . Myers, M. J. Kim, M. Du, A. Lee, K. Fang, C. Finn, and S. Levine, “BridgeData V2: A dataset for robot learning at scale,” inProceedings of The 7th Conference on Robot Learning, ser. Proceed...

    2023

  8. [16]

    Vision-language foundation models as effective robot imitators,

    X. Li, M. Liu, H. Zhang, C. Yu, J. Xu, H. Wu, C. Cheang, Y . Jing, W. Zhang, H. Liu, H. Li, and T. Kong, “Vision-language foundation models as effective robot imitators,” inThe Twelfth International Conference on Learning Representations, 2024. [Online]. Available: https://ope...

    2024

  9. [17]

    Visa-flow: Accelerating robot skill learning via large-scale video semantic action flow,

    C. Chen, Q. Yang, X. Xu, N. Fazeli, and O. Andersson, “Visa-flow: Accelerating robot skill learning via large-scale video semantic action flow,”arXiv preprint arXiv:2505.01288, 2025

    2025

  10. [18]

    π 0: A vision-language-action flow model for general robot control,

    K. Black, N. Brown, D. Driess, A. Esmail, M. R. Equi, C. Finn, N. Fusai, L. Groom, K. Hausman, B. Ichter, S. Jakubczak, T. Jones, L. Ke, S. Levine, A. Li-Bell, M. Mothukuri, S. Nair, K. Pertsch, L. X. Shi, L. Smith, J. Tanner, Q. Vuong, A. Walling, H. Wang, and U. Zhilinsky, “...

    2025

  11. [19]

    π0.5: A vision-language-action model with open-world generalization,

    K. Black, N. Brown, J. Darpinian, K. Dhabalia, D. Driess, A. Esmail, M. R. Equi, C. Finn, N. Fusai, M. Y . Galliker, D. Ghosh, L. Groom, K. Hausman, B. Ichter, S. Jakubczak, T. Jones, L. Ke, D. LeBlanc, S. Levine, A. Li-Bell, M. Mothukuri, S. Nair, K. Pertsch, A. Z. Ren, L. X....

  12. [20]

    GR00T N1: An open foundation model for generalist humanoid robots,

    NVIDIA, J. Bjorck, F. Castañeda, N. Cherniadev, X. Da, R. Ding, L. Fan, Y . Fang, D. Fox, F. Hu, S. Huang, J. Jang, Z. Jiang, J. Kautz, K. Kundalia, L. Lao, Z. Li, Z. Lin, K. Lin, G. Liu, E. Llontop, L. Magne, A. Mandlekar, A. Narayan, S. Nasiriany, S. Reed, Y . L. Tan, G. Wan...

    2025 arXiv

  13. [21]

    Interactive post- training for vision-language-action models,

    S. Tan, K. Dou, Y . Zhao, and P. Krähenbühl, “Interactive post- training for vision-language-action models,” 2025. [Online]. Available: https://arxiv.org/abs/2505.17016

    2025 arXiv

  14. [22]

    VLA-RFT: Vision-language-action reinforcement fine-tuning with verified rewards in world simulators,

    H. Li, P. Ding, R. Suo, Y . Wang, Z. Ge, D. Zang, K. Yu, M. Sun, H. Zhang, D. Wang, and W. Su, “VLA-RFT: Vision-language-action reinforcement fine-tuning with verified rewards in world simulators,”

  15. [23]

    ConRFT: A reinforced fine-tuning method for VLA models via consistency policy,

    Y . Chen, S. Tian, S. Liu, Y . Zhou, H. Li, and D. Zhao, “ConRFT: A reinforced fine-tuning method for VLA models via consistency policy,” inProceedings of Robotics: Science and Systems, Los Angeles, CA, USA, June 2025. [Online]. Available: https://www.roboticsproceedings.org/r...

    2025

  16. [24]

    π ∗ 0.6: a VLA that learns from experience,

    Physical Intelligence, A. Amin, R. Aniceto, A. Balakrishna, K. Black, K. Conley, G. Connors, J. Darpinian, K. Dhabalia, J. DiCarlo, D. Driess, M. Equi, A. Esmail, Y . Fang, C. Finn, C. Glossop, T. Godden, I. Goryachev, L. Groom, H. Hancock, K. Hausman, G. Hussein, B. Ichter, S...

    2025 arXiv

  17. [25]

    Acting while understanding: Asynchronous semantic-action decoupling for real-time vision-language-action models,

    S. Yan, G. Wang, Q. Liu, W. Meng, J. Yang, C. Yao, F. Feng, X. Ma, Y . Zhao, and Y . Han, “Acting while understanding: Asynchronous semantic-action decoupling for real-time vision-language-action models,” 2026. [Online]. Available: https://arxiv.org/abs/2606.15285

    2026

  18. [26]

    DAM-VLA: Decoupled asynchronous multimodal vision language action model,

    P. Vanjani, Z. Li, J. Suliga, M. Reuss, G. Geraci, X. Jiang, and R. Lioutikov, “DAM-VLA: Decoupled asynchronous multimodal vision language action model,” 2026. [Online]. Available: https: //arxiv.org/abs/2606.12105

    2026 arXiv

  19. [27]

    RL-VLA 3: A flexible and asynchronous reinforcement learning framework for VLA training,

    H. Sun, Y . Guo, Z. Guan, S. Di, X. Bai, J. Long, T. Zhao, M. Luo, H. Zhao, L. Wu, X. Deng, X. Chu, X. Xiao, S. Wen, Y . Gong, and J. Xiong, “RL-VLA 3: A flexible and asynchronous reinforcement learning framework for VLA training,” 2026. [Online]. Available: https://arxiv.org/...

    2026 arXiv

  20. [28]

    FLOWER: Democratizing generalist robot policies with efficient vision-language-flow models,

    M. Reuss, H. Zhou, M. Rühle, Ö. E. Ya ˘gmurlu, F. Otto, and R. Lioutikov, “FLOWER: Democratizing generalist robot policies with efficient vision-language-flow models,” inProceedings of The 9th Conference on Robot Learning, ser. Proceedings of Machine Learning Research, vol. 30...

    2025

  21. [29]

    CALVIN: A benchmark for language-conditioned policy learning for long-horizon robot manipulation tasks,

    O. Mees, L. Hermann, E. Rosete-Beas, and W. Burgard, “CALVIN: A benchmark for language-conditioned policy learning for long-horizon robot manipulation tasks,”IEEE Robotics and Automation Letters, vol. 7, no. 3, pp. 7327–7334, 2022

    2022

  22. [30]

    Unleashing large-scale video generative pre-training for visual robot manipulation,

    H. Wu, Y . Jing, C. Cheang, G. Chen, J. Xu, X. Li, M. Liu, H. Li, and T. Kong, “Unleashing large-scale video generative pre-training for visual robot manipulation,” 2023. [Online]. Available: https://arxiv.org/abs/2312.13139

    2023 arXiv

  23. [31]

    Unified vision-language-action model,

    Y . Wang, X. Li, W. Wang, J. Zhang, Y . Li, Y . Chen, X. Wang, and Z. Zhang, “Unified vision-language-action model,” 2025. [Online]. Available: https://arxiv.org/abs/2506.19850

    2025 arXiv

  24. [32]

    Disentangled robot learning via separate forward and inverse dynamics pretraining,

    W. Zhang, B. Zhang, Z. Qi, W. Zeng, X. Jin, and L. Zhang, “Disentangled robot learning via separate forward and inverse dynamics pretraining,” inThe Fourteenth International Conference on Learning Representations, 2026. [Online]. Available: https: //openreview.net/forum?id=DdrsHWobR1

    2026

  25. [33]

    Zero-shot robotic manipulation with pretrained image-editing diffusion models,

    K. Black, M. Nakamoto, P. Atreya, H. Walke, C. Finn, A. Kumar, and S. Levine, “Zero-shot robotic manipulation with pretrained image-editing diffusion models,” 2023. [Online]. Available: https: //arxiv.org/abs/2310.10639

    2023 arXiv

  26. [34]

    3D diffuser actor: Policy diffusion with 3D scene representations,

    T.-W. Ke, N. Gkanatsios, and K. Fragkiadaki, “3D diffuser actor: Policy diffusion with 3D scene representations,” 2024. [Online]. Available: https://arxiv.org/abs/2402.10885

    2024 arXiv

  27. [35]

    Closed-loop visuomotor control with generative expectation for robotic manipulation,

    Q. Bu, J. Zeng, L. Chen, Y . Yang, G. Zhou, J. Yan, P. Luo, H. Cui, Y . Ma, and H. Li, “Closed-loop visuomotor control with generative expectation for robotic manipulation,” inAdvances in Neural Infor- mation Processing Systems, vol. 37, 2024, pp. 139 002–139 029

    2024

  28. [36]

    Towards synergistic, generalized, and efficient dual-system for robotic manipulation,

    Q. Bu, H. Li, L. Chen, J. Cai, J. Zeng, H. Cui, M. Yao, and Y . Qiao, “Towards synergistic, generalized, and efficient dual-system for robotic manipulation,” 2024. [Online]. Available: https://arxiv.org/abs/2410.08001

    2024 arXiv

  29. [37]

    UP-VLA: A unified understanding and prediction model for embodied agent,

    J. Zhang, Y . Guo, Y . Hu, X. Chen, X. Zhu, and J. Chen, “UP-VLA: A unified understanding and prediction model for embodied agent,”

  30. [38]

    What matters in building vision-language- action models for generalist robots,

    X. Li, P. Li, L. Qian, M. Liu, D. Wang, J. Liu, B. Kang, X. Ma, X. Wang, D. Guo,et al., “What matters in building vision-language- action models for generalist robots,”Nature Machine Intelligence, pp. 1–15, 2026

    2026

  31. [39]

    Predictive inverse dynamics models are scalable learners for robotic manipulation,

    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,” 2024. [Online]. Available: https://arxiv.org/abs/2412.15109

    2024 arXiv

  32. [40]

    Available: https://arxiv.org/abs/2501.18867

    [Online]. Available: https://arxiv.org/abs/2501.18867

    arXiv

  33. [43]

    Video prediction policy: A generalist robot policy with predictive visual representations,

    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,” inProceedings of the 42nd International Conference on Machine Learning, 2025. [Online]. Avail...

    2025

  34. [305]

    PMLR, 27–30 Sep 2025, pp. 17–40. [Online]. Available: https://proceedings.mlr.press/v305/black25a.html

    2025

  35. [2025]

    Available: https://arxiv.org/abs/2510.00406

    [Online]. Available: https://arxiv.org/abs/2510.00406

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.