Pith. sign in

REVIEW 3 major objections 6 minor 51 references

Steering robot policies with primitive-level guidance at inference time

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 · glm-5.2

2026-07-09 20:28 UTC pith:BYPUGGLI

load-bearing objection Plug-and-play test-time primitive guidance for diffusion/flow policies: real gains, but LIBERO results lack seed variance and the error bound is unverified. the 3 major comments →

arxiv 2607.07076 v1 pith:BYPUGGLI submitted 2026-07-08 cs.RO

PriGo: Test-Time Primitive Guidance to Diffusion and Flow Policies for Adaptive Robotic Manipulation

Zezeng Li , Enda Xiang , Thuy Tran , Di Huang , Momath Thiam , Liming Chen This is my paper
classification cs.RO
keywords policiesprigodiffusionprimitiveroboticbehaviorsmanipulationacross
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.

Robot imitation policies that generate actions via diffusion or flow matching can produce motions that look locally plausible but violate the structural intent of a manipulation task — for example, pulling before rotating a key, or grasping the wrong part of a cup. This paper claims that a lightweight classifier, trained separately to predict which of eight motion primitives (grasp, push, pull, rotate, etc.) the robot should be executing at each step, can be used as a differentiable constraint during inference to nudge the policy's output actions toward structurally consistent trajectories. The key claim is that this guidance requires no retraining of the underlying policy: at each denoising or flow-matching step, the gradient of a cross-entropy loss between the classifier's predicted primitive distribution and a soft classification of the current action proposal is used to refine the action. The authors report consistent 3–7 point success-rate improvements across four simulation benchmarks and real-world table-service tasks, for both diffusion-based and flow-based policies of varying scale.

Core claim

The central mechanism is a test-time correction loop. A pretrained primitive classifier (PANet) maps the current visual observation and language instruction to a probability distribution over eight manipulation primitives. Separately, each candidate action generated by a frozen diffusion or flow policy is converted into a soft probability distribution over the same eight categories using differentiable score functions on the action's translational, rotational, and gripper-width components. The cross-entropy between these two distributions serves as a guidance loss, and its gradient with respect to the action is applied as a correction term at each inference step — modifying the denoising or,

What carries the argument

The differentiable primitive guidance mechanism. The hard primitive classification (Eq. 1) maps a 7-DOF action to one of eight categories using thresholds on translation magnitude, rotation magnitude, and gripper-width change. The soft classification (Def. 3.2) replaces the argmax with a temperature-scaled softmax over differentiable score functions, making the mapping from action to primitive distribution smooth. The guidance loss L_PG (Eq. 2) is the cross-entropy between PANet's predicted distribution and the action-induced soft distribution. For diffusion policies, the gradient ∇_a L_PG is applied to the estimated clean action at each denoising step (Eq. 3); for flow policies, it modifies

Load-bearing premise

The eight-category primitive taxonomy and its fixed classification thresholds capture manipulation structure that is both predictable from visual observations by a small classifier and useful as a gradient-based constraint for action refinement. If the taxonomy is too coarse, too fine, or misaligned with the task's actual motion structure, the guidance gradient could be uninformative or counterproductive.

What would settle it

A manipulation domain where the eight primitives do not decompose the relevant actions — for example, tasks dominated by continuous deformations (folding cloth, pouring liquid) or tasks where the same observation legitimately requires different primitives depending on hidden state. In such domains, PANet's predictions would be unreliable, and the guidance gradient would push actions toward the wrong structural category, degrading rather than improving policy performance.

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

If this is right

  • Any policy that generates actions as iterative refinement of noisy samples (diffusion, flow matching, or related generative processes) could in principle receive test-time guidance from an external classifier that predicts a structurally meaningful property of the action, not just primitives — for example, contact state, affordance, or object-relative pose.
  • The plug-and-play design means that organizations deploying pretrained manipulation policies could improve robustness without access to the original training data or policy weights, applying only an inference-time gradient correction.
  • The primitive taxonomy and its automatic labeling procedure (Eq. 1) could be extended or adapted to new manipulation domains by redefining the primitive set and retraining only the lightweight classifier, leaving the policy untouched.
  • The error-bound analysis (Proposition A.1) suggests that guidance fields with a fixed point at the target action and local strong monotonicity guarantee convergence improvement, providing a theoretical template for analyzing other test-time guidance schemes.

Where Pith is reading between the lines

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

  • If the primitive taxonomy is domain-specific and manually designed, the approach may transfer poorly to manipulation domains with fundamentally different action structures — for instance, in-hand manipulation or deformable-object manipulation where the relevant primitives (e.g., pivoting, rolling, stretching) differ from the eight proposed here.
  • The guidance gradient could interact poorly with the policy's learned distribution in cases where the policy intentionally produces actions that do not match any single primitive — for example, smooth transitional motions or bimanual coordinated actions that blend primitives simultaneously. The soft classification mitigates this, but the taxonomy's discreteness may still introduce bias.
  • A natural extension would be to learn the primitive taxonomy itself from data rather than designing it manually, potentially via clustering in action-embedding space, which could make the guidance more adaptive to unseen task distributions.
  • The 15% inference-time overhead reported (181 ms → 208 ms per step) is modest for the benchmarks tested, but the overhead scales with the number of denoising or flow steps and the frequency of guidance application, which could become significant for policies with many refinement steps or high-frequency 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

3 major / 6 minor

Summary. PriGo introduces a test-time primitive guidance framework for diffusion and flow policies in robotic manipulation. The core idea is to train a lightweight classifier (PANet) that predicts an 8-category primitive distribution from visual observations and language instructions, then use a differentiable guidance loss (cross-entropy between PANet predictions and soft-classified actions) to refine generated actions during inference via gradient correction. The guidance is integrated into both DDPM-based diffusion policies and flow matching policies without retraining the backbone. Experiments span LIBERO, CALVIN, SIMPLER, and real-world Franka tasks across four backbones (π0, SmolVLA, DP, CogACT), with ablations on the primitive taxonomy and PANet accuracy.

Significance. The plug-and-play, no-retraining design is the paper's main contribution and is well-motivated: it decouples primitive planning from continuous action learning and applies to both lightweight and VLA-scale policies. The differentiable soft-classification (Def. 3.2) enabling gradient flow through the primitive taxonomy is a clean technical contribution. The error-bound analysis (Proposition A.1) attempts to formalize convergence, and the sensitivity analysis (Fig. 3, right) addresses threshold robustness. The breadth of benchmarks and the real-world demonstrations with limited data are commendable. Code is promised, which supports reproducibility.

major comments (3)
  1. Table 1 (LIBERO): The headline claim that PriGo 'consistently improves' performance by 3–5 points is reported without seed-level variance. Diffusion and flow policies are well-documented to exhibit ±3–5 point run-to-run variance from stochastic initialization, data ordering, and inference-time noise sampling. Several gains in Table 1 (e.g., π0 Spatial: +1.8, SmolVLA Goal: +1.8, DP Goal: +2.2) are within this band. Without confidence intervals or multi-seed evaluation on the primary benchmark, the word 'consistently' is unsupported for the settings where gains are smallest. The real-world (Table 4: +22–29 pts) and CALVIN (Table 3: +0.83 avg length) results show larger margins and are more convincing, but LIBERO is the in-domain benchmark where the smallest gains are reported. Please provide multi-seed results (≥3 seeds) with standard deviations for Table 1, or reframe the claim to match.
  2. §3.2–3.3: The guidance step size η in Eqs. (3) and (4) is never specified anywhere in the paper, and no sensitivity analysis is provided. This is a load-bearing hyperparameter: if η is too small, the guidance is negligible and the improvements are unrelated to the proposed mechanism; if too large, it may destabilize trajectories. The sensitivity analysis in Fig. 3 (right) covers only the auto-labeling thresholds (τ_trans, τ_rot, τ_w), not η or the softmax temperature τ in Def. 3.2. Please report the η values used for each backbone and provide a sensitivity analysis, as this directly determines whether the guidance mechanism is actually responsible for the observed gains.
  3. Proposition A.1 (Appendix A): The error bound assumes the guidance field G is locally strongly monotone near the target action (Eq. 9: ⟨G(a)−G(a*), a−a*⟩ ≥ C₀‖a−a*‖²). This is a strong assumption that is not verified for the cross-entropy-based guidance loss L_PG. The guidance gradient ∇_a L_PG depends on the soft-classification score functions f_k (Def. 3.2), whose monotonicity properties near arbitrary target actions are not established. The proposition thus provides a formal bound under an assumption whose applicability is unclear. Please either verify the monotonicity condition empirically (e.g., by plotting ‖∇_a L_PG‖ vs. ‖a−a*‖ near expert actions) or clearly state this as an unverified assumption and soften the theoretical contribution accordingly.
minor comments (6)
  1. §3.1, Eq. (1): The hard classification uses x_k^1 (the first component of translation) to distinguish push (x_k^1 > 0) from pull (x_k^1 < 0). It is unclear which axis this corresponds to (robot base frame? end-effector frame? world frame?). Please specify, as the push/pull distinction depends on this choice.
  2. Table 2: The caption states 'Comparison of our approach PriGo-DP (CogACT with PriGo)' but the table header lists 'PriGo-DP' as a separate method row alongside CogACT. This is slightly confusing—consider labeling it 'CogACT + PriGo' for consistency with Table 1's notation.
  3. §4.3: The real-world experiments report 'averaged success rates over 10 episodes per task' (Table 4). While the gains are large (+22–29 pts), 10 episodes is a small sample. Please report confidence intervals or note this limitation.
  4. Appendix C: The wall-clock analysis (208 ms/step, 77 Hz effective) is mentioned only in the appendix. Given that inference overhead is practically important for a plug-and-play method, consider moving this to the main text.
  5. Fig. 1: The figure caption references 'a) Structured Primitive Actions' and 'b) Unstructured Primitive Actions,' but the subfigure labels in the image appear to show specific failure modes (pull before unlock, falling) rather than a general structured vs. unstructured comparison. Consider revising the caption for clarity.
  6. The manuscript states 'Codes are available on PriGo' (§1) but no URL is provided. Please include a repository link or anonymized placeholder.

Circularity Check

0 steps flagged

No circularity: primitive guidance loss is observation-conditioned, not self-referential

full rationale

The paper's central mechanism is: (1) PANet predicts primitive distributions from observations O (Eq. 2: L_PG = H(PANet(O_k), p_k)), and (2) the guidance gradient ∇_a L_PG refines generated actions during inference (Eqs. 3-4). The primitive labels for training PANet are auto-generated from action properties via hard classification (Eq. 1: thresholds on translation, rotation, gripper width). This is not circular because PANet's input (observations) is distinct from the guided quantity (actions). The guidance loss compares PANet's observation-conditioned prediction against the soft-classified current action — if the action already matches the predicted primitive, the gradient is zero; if it deviates, the gradient pushes it back. The prediction is not defined in terms of the action being guided. The error bound (Proposition A.1) assumes G(a*) = 0 (target action is a fixed point of guidance) and local strong monotonicity, then derives a contraction — this is a standard convergence argument, not a circular definition. The taxonomy design (8 primitives) is validated empirically (Table 5 ablation, Appendix B distribution analysis), not by self-citation. No self-citation chain is load-bearing for the central claim. The derivation is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

3 free parameters · 3 axioms · 2 invented entities

Legacy field; see axiom_ledger.

free parameters (3)
  • τ_trans, τ_rot, τ_w = Not explicitly stated; sensitivity analysis in Fig. 3 shows stable performance across a range
    Thresholds for hard primitive classification (Eq. 1). Determine which primitive category an action belongs to. Sensitivity analysis shows robustness but specific values are not reported.
  • η (guidance step size) = Not explicitly stated
    Controls the strength of primitive guidance in Eqs. 3–4. Affects how aggressively actions are refined toward primitive-consistent trajectories.
  • τ (softmax temperature) = Not explicitly stated
    Temperature for soft classification (Def. 3.2). Controls the sharpness of the primitive probability distribution induced by actions.
axioms (3)
  • domain assumption Manipulation behaviors decompose into a small set of reusable motion primitives
    Section 1: 'Human manipulation behaviors naturally exhibit compositional structure, where complex tasks are composed of a small number of reusable motion primitives.' This motivates the entire framework.
  • domain assumption The 8-category primitive taxonomy (idle, grasp, release, push, pull, rotation, push+rotation, pull+rotation) is sufficient for common manipulation tasks
    Section 3.1: 'We adopt this compact primitive taxonomy to balance structural expressiveness and cross-task generalization.' Validated empirically on tested benchmarks but not proven complete.
  • ad hoc to paper The guidance field G is locally strongly monotone near target actions
    Appendix A, Eq. 9: Required for the error bound (Proposition A.1). Not verified empirically; stated as an assumption for the proof.
invented entities (2)
  • PANet (Primitive Action Network) independent evidence
    purpose: Predicts primitive distributions from visual observations and language instructions
    Trained with auto-generated labels (Eq. 1), achieves 94% classification accuracy on LIBERO (Fig. 3). Confusion matrix shows errors are mostly between semantically adjacent primitives. Ablation in Table 5 confirms contribution of each primitive category.
  • 8-category primitive taxonomy independent evidence
    purpose: Provides structured action priors for guidance
    Ablation study (Table 5) compares full 8-primitive set against reduced sets, showing performance degradation when categories are removed. Distribution analysis in Appendix B justifies the choice over finer-grained taxonomies.

pith-pipeline@v1.1.0-glm · 17907 in / 3322 out tokens · 104166 ms · 2026-07-09T20:28:20.351758+00:00 · methodology

0 comments
read the original abstract

Imitation learning has enabled remarkable progress in robotic manipulation, especially with diffusion and flow-based policies that generate complex visuomotor behaviors directly from demonstrations. Yet, despite their strong performance, these policies often fail to generalize across tasks and environments. A key reason is that existing policies tend to imitate superficial action correlations rather than the underlying intent. Inspired by the compositional structure of human behaviors, we propose PriGo, a primitive-guided test-time adaptive framework for robust robotic manipulation. PriGo introduces PANet, a lightweight primitive prediction module that infers primitive distributions directly from observations. We further propose a differentiable primitive guidance mechanism that refines generated actions during inference, steering trajectories toward semantically consistent behaviors. Unlike prior primitive-conditioned approaches, PriGo operates entirely at test time and can be seamlessly integrated into pretrained diffusion and flow policies without retraining. Extensive experiments on LIBERO, CALVIN, SIMPLER, and real-world robotic tasks demonstrate that PriGo consistently improves robustness, long-horizon execution, and generalization ability across both diffusion and flow-based policies.

Figures

Figures reproduced from arXiv: 2607.07076 by Di Huang, Enda Xiang, Liming Chen, Momath Thiam, Thuy Tran, Zezeng Li.

Figure 1
Figure 1. Figure 1: Comparison of a) our policy with PriGo, which produces structured primitive actions, and b) vanilla policies without PriGo, which lack structured primitive action outputs. We then incorporate differentiable primitive guidance into both diffusion and flow policies. During inference, actions are refined using gradients derived from primitive consistency objectives, encour￾aging trajectories to align with the… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of PriGo, a primitive-guided framework for robotic manipulation policies. Given visual observations and language instructions, the primitive classifier PANet predicts probabilistic manipulation primitives. During inference, generated actions from pretrained diffusion or flow policies are converted into soft primitive distributions and aligned with the predicted primitives through differentiable pr… view at source ↗
Figure 3
Figure 3. Figure 3: Confusion matrix of PANet primitive classification (left) and sensitivity analysis of auto [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Rollouts of our PriGo-DP on three table-service tasks. PriGo-DP aligns the generated [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Visualization of actions and primitive labels from two episodes via UMAP and t-SNE. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of our PriGo-DP (top row) and the vanilla diffusion policy (bottom row) on [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗

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

51 extracted references · 51 canonical work pages · 11 internal anchors

  1. [1]

    Sapkota, Y

    R. Sapkota, Y . Cao, K. I. Roumeliotis, and M. Karkee. Vision-language-action models: Concepts, progress, applications and challenges.arXiv preprint arXiv:2505.04769, 2025

    work page arXiv 2025

  2. [2]

    Z. Li, A. Chapin, E. Xiang, R. Yang, B. Machado, N. Lei, E. Dellandrea, D. Huang, and L. Chen. Robotic manipulation via imitation learning: Taxonomy, evolution, benchmark, and challenges. arXiv preprint arXiv:2508.17449, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  3. [3]

    Hiranaka, M

    A. Hiranaka, M. Hwang, S. Lee, C. Wang, L. Fei-Fei, J. Wu, and R. Zhang. Primitive skill-based robot learning from human evaluative feedback. In2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 7817–7824. IEEE, 2023

    work page 2023

  4. [4]

    T. Gao, S. Nasiriany, H. Liu, Q. Yang, and Y . Zhu. Prime: Scaffolding manipulation tasks with behavior primitives for data-efficient imitation learning.IEEE Robotics and Automation Letters, 2024

    work page 2024

  5. [5]

    C. Chi, Z. Xu, S. Feng, E. Cousineau, Y . Du, B. Burchfiel, R. Tedrake, and S. Song. Diffusion policy: Visuomotor policy learning via action diffusion.The International Journal of Robotics Research, page 02783649241273668, 2023

    work page 2023

  6. [6]

    Black, M

    K. Black, M. Nakamoto, P. Atreya, H. R. Walke, C. Finn, A. Kumar, and S. Levine. Zero- shot robotic manipulation with pre-trained image-editing diffusion models. InThe Twelfth International Conference on Learning Representations, 2024

    work page 2024

  7. [7]

    Z. Xian, N. Gkanatsios, T. Gervet, T.-W. Ke, and K. Fragkiadaki. Chaineddiffuser: Unifying trajectory diffusion and keypose prediction for robotic manipulation. InConference on Robot Learning, pages 2323–2339. PMLR, 2023

    work page 2023

  8. [8]

    Y . Wang, X. Zhang, R. Wu, Y . Li, Y . Shen, M. Wu, Z. He, Y . Wang, and H. Dong. Adamanip: Adaptive articulated object manipulation environments and policy learning. InThe Thirteenth International Conference on Learning Representations, 2025

    work page 2025

  9. [9]

    Y . Ze, G. Zhang, K. Zhang, C. Hu, M. Wang, and H. Xu. 3d diffusion policy: Generalizable visuomotor policy learning via simple 3d representations. InICRA 2024 Workshop on 3D Visual Representations for Robot Manipulation, 2024

    work page 2024

  10. [10]

    J. Yang, Z. Cao, C. Deng, R. Antonova, S. Song, and J. Bohg. Equibot: Sim (3)-equivariant diffusion policy for generalizable and data efficient learning. In8th Annual Conference on Robot Learning, 2024

    work page 2024

  11. [11]

    T.-W. Ke, N. Gkanatsios, and K. Fragkiadaki. 3d diffuser actor: Policy diffusion with 3d scene representations. In8th Annual Conference on Robot Learning, 2024

    work page 2024

  12. [12]

    D. Wang, S. Hart, D. Surovik, T. Kelestemur, H. Huang, H. Zhao, M. Yeatman, J. Wang, R. Walters, and R. Platt. Equivariant diffusion policy.8th Annual Conference on Robot Learning, 2024

    work page 2024

  13. [13]

    H. He, C. Bai, L. Pan, W. Zhang, B. Zhao, and X. Li. Learning an actionable discrete diffusion policy via large-scale actionless video pre-training.Advances in Neural Information Processing Systems, 37:31124–31153, 2025. 9

    work page 2025

  14. [14]

    Xiang, H

    E. Xiang, H. Ma, X. Ma, Z. Liu, and D. Huang. Graspldp: Towards generalizable grasping policy via latent diffusion.arXiv preprint arXiv:2602.22862, 2026

    work page arXiv 2026

  15. [15]

    O. M. Team, D. Ghosh, H. Walke, K. Pertsch, K. Black, O. Mees, S. Dasari, J. Hejna, T. Kreiman, C. Xu, et al. Octo: An open-source generalist robot policy.arXiv:2405.12213, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  16. [16]

    J. Wen, M. Zhu, Y . Zhu, Z. Tang, J. Li, Z. Zhou, C. Li, X. Liu, Y . Peng, C. Shen, et al. Diffusion-vla: Scaling robot foundation models via unified diffusion and autoregression. arXiv:2412.03293, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  17. [17]

    Q. Li, Y . Liang, Z. Wang, L. Luo, X. Chen, M. Liao, F. Wei, et al. Cogact: A foundational vision-language-action model for synergizing cognition and action in robotic manipulation. arXiv:2411.19650, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  18. [18]

    Z. Zhou, Y . Zhu, M. Zhu, J. Wen, N. Liu, Z. Xu, W. Meng, Y . Peng, C. Shen, F. Feng, et al. Chatvla: Unified multimodal understanding and robot control with vision-language-action model. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 5377–5395, 2025

    work page 2025

  19. [19]

    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:2410.07864, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  20. [20]

    J. Liu, H. Chen, P. An, Z. Liu, R. Zhang, C. Gu, X. Li, Z. Guo, S. Chen, M. Liu, et al. Hybridvla: Collaborative diffusion and autoregression in a unified vision-language-action model.arXiv:2503.10631, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  21. [21]

    Rouxel, A

    Q. Rouxel, A. Ferrari, S. Ivaldi, and J.-B. Mouret. Flow matching imitation learning for multi- support manipulation. In2024 IEEE-RAS 23rd International Conference on Humanoid Robots (Humanoids), pages 528–535. IEEE, 2024

    work page 2024

  22. [22]

    Zhang and M

    F. Zhang and M. Gienger. Affordance-based robot manipulation with flow matching.arXiv preprint arXiv:2409.01083, 2024

    work page arXiv 2024

  23. [23]

    N. Funk, J. Urain, J. Carvalho, V . Prasad, G. Chalvatzaki, and J. Peters. Actionflow: Effi- cient, accurate, and fast policies with spatially symmetric flow matching. InR: SS workshop: Structural Priors as Inductive Biases for Learning Robot Dynamics, 2024

    work page 2024

  24. [24]

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

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

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  25. [25]

    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

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  26. [26]

    S. Deng, M. Yan, S. Wei, H. Ma, Y . Yang, J. Chen, Z. Zhang, T. Yang, X. Zhang, H. Cui, et al. Graspvla: a grasping foundation model pre-trained on billion-scale synthetic action data.arXiv preprint arXiv:2505.03233, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  27. [27]

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

    M. Shukor, D. Aubakirova, F. Capuano, P. Kooijmans, S. Palma, A. Zouitine, M. Aractingi, C. Pascal, M. Russi, A. Marafioti, et al. Smolvla: A vision-language-action model for affordable and efficient robotics.arXiv:2506.01844, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  28. [28]

    F. Ni, J. Hao, Y . Mu, Y . Yuan, Y . Zheng, B. Wang, and Z. Liang. Metadiffuser: Diffusion model as conditional planner for offline meta-rl. InInternational Conference on Machine Learning, pages 26087–26105. PMLR, 2023

    work page 2023

  29. [29]

    W. K. Kim, Y . Lee, J. Kim, and H. Woo. Llm-based skill diffusion for zero-shot policy adaptation.Advances in Neural Information Processing Systems, 37:6749–6775, 2024. 10

    work page 2024

  30. [30]

    Z. Li, R. Yang, R. Chen, Z. Luo, and L. Chen. Adpro: a test-time adaptive diffusion policy for robot manipulation via manifold and initial noise constraints.arXiv preprint arXiv:2508.06266, 2025

    work page arXiv 2025

  31. [31]

    X. B. Peng, M. Chang, G. Zhang, P. Abbeel, and S. Levine. Mcp: Learning composable hierarchical control with multiplicative compositional policies.Advances in neural information processing systems, 32, 2019

    work page 2019

  32. [32]

    Merel, L

    J. Merel, L. Hasenclever, A. Galashov, A. Ahuja, V . Pham, G. Wayne, Y . Teh, and N. Heess. Neural probabilistic motor primitives for humanoid control. InInternational Conference on Learning Representations, 2019

    work page 2019

  33. [33]

    Knaust and D

    M. Knaust and D. Koert. Guided robot skill learning: A user-study on learning probabilistic movement primitives with non-experts. In2020 IEEE-RAS 20th International Conference on Humanoid Robots (Humanoids), pages 514–521. IEEE, 2021

    work page 2021

  34. [34]

    Dalal, D

    M. Dalal, D. Pathak, and R. R. Salakhutdinov. Accelerating robotic reinforcement learning via parameterized action primitives.Advances in Neural Information Processing Systems, 34: 21847–21859, 2021

    work page 2021

  35. [35]

    Zhang, J

    J. Zhang, J. Zhang, K. Pertsch, Z. Liu, X. Ren, M. Chang, S.-H. Sun, and J. J. Lim. Bootstrap your own skills: Learning to solve new tasks with large language model guidance. In7th Annual Conference on Robot Learning, 2023

    work page 2023

  36. [36]

    Garcia, S

    R. Garcia, S. Chen, and C. Schmid. Towards generalizable vision-language robotic manipulation: A benchmark and llm-guided 3d policy. In2025 IEEE International Conference on Robotics and Automation (ICRA), pages 8996–9002. IEEE, 2025

    work page 2025

  37. [37]

    M. Pan, J. Zhang, T. Wu, Y . Zhao, W. Gao, and H. Dong. Omnimanip: Towards general robotic manipulation via object-centric interaction primitives as spatial constraints. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 17359–17369, 2025

    work page 2025

  38. [38]

    Q. Sun, L. Yang, W. Tang, W. Huang, K. Xu, Y . Chen, M. Liu, J. Yang, H. Zhu, Y . Wang, et al. Learning primitive embodied world models: Towards scalable robotic learning.arXiv preprint arXiv:2508.20840, 2025

    work page arXiv 2025

  39. [39]

    T. Liu, J. Li, Y . Zheng, H. Niu, Y . Lan, X. Xu, and X. Zhan. Skill expansion and composition in parameter space. InThe Thirteenth International Conference on Learning Representations, 2025

    work page 2025

  40. [40]

    W. Wan, Y . Zhu, R. Shah, and Y . Zhu. Lotus: Continual imitation learning for robot manipulation through unsupervised skill discovery. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 537–544, 2024

    work page 2024

  41. [41]

    Y . Guo, B. Tang, I. Akinola, D. Fox, A. Gupta, and Y . Narang. Srsa: Skill retrieval and adaptation for robotic assembly tasks. InThe Thirteenth International Conference on Learning Representations, 2025

    work page 2025

  42. [42]

    Oquab, T

    M. Oquab, T. Darcet, T. Moutakanni, H. V o, M. Szafraniec, V . Khalidov, P. Fernandez, D. Haziza, F. Massa, A. El-Nouby, et al. Dinov2: Learning robust visual features without supervision. Transactions on Machine Learning Research Journal, pages 1–31, 2024

    work page 2024

  43. [43]

    Raffel, N

    C. Raffel, N. Shazeer, A. Roberts, K. Lee, S. Narang, M. Matena, Y . Zhou, W. Li, and P. J. Liu. Exploring the limits of transfer learning with a unified text-to-text transformer.Journal of Machine Learning Research, 21(140):1–67, 2020. URL http://jmlr.org/papers/v21/ 20-074.html

    work page 2020

  44. [44]

    J. Ho, A. Jain, and P. Abbeel. Denoising diffusion probabilistic models. InAdvances in Neural Information Processing Systems, volume 33, pages 6840–6851, 2020. 11

    work page 2020

  45. [45]

    B. Liu, Y . Zhu, C. Gao, Y . Feng, Q. Liu, Y . Zhu, and P. Stone. Libero: Benchmarking knowledge transfer for lifelong robot learning.Advances in Neural Information Processing Systems, 36: 44776–44791, 2023

    work page 2023

  46. [46]

    X. Li, K. Hsu, J. Gu, O. Mees, K. Pertsch, H. R. Walke, C. Fu, I. Lunawat, I. Sieh, S. Kirmani, et al. Evaluating real-world robot manipulation policies in simulation. InConference on Robot Learning, pages 3705–3728. PMLR, 2025

    work page 2025

  47. [47]

    Vuong, S

    Q. Vuong, S. Levine, H. R. Walke, K. Pertsch, A. Singh, R. Doshi, C. Xu, J. Luo, L. Tan, D. Shah, et al. Open x-embodiment: Robotic learning datasets and rt-x models. InTowards Generalist Robots: Learning Paradigms for Scalable Skill Acquisition@ CoRL2023, 2023

    work page 2023

  48. [48]

    M. J. Kim, K. Pertsch, S. Karamcheti, T. Xiao, A. Balakrishna, S. Nair, R. Rafailov, E. P. Foster, P. R. Sanketi, Q. Vuong, et al. Openvla: An open-source vision-language-action model. In Conference on Robot Learning, pages 2679–2713. PMLR, 2025

    work page 2025

  49. [49]

    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, 7(3):7327–7334, 2022

    work page 2022

  50. [50]

    Z. Hou, T. Zhang, Y . Xiong, H. Pu, C. Zhao, R. Tong, Y . Qiao, J. Dai, and Y . Chen. Diffusion transformer policy.arXiv preprint arXiv:2410.15959, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  51. [51]

    pick up the salad dressing and place it in the basket

    J. Li, D. Li, S. Savarese, and S. Hoi. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. InInternational conference on machine learning, pages 19730–19742. PMLR, 2023. 12 This document presents the error-bound proposition, along with implementation details for PANet and a primitive-action analysis. It ...

    work page 2023