Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

ActionCache: Training-Free Acceleration for Vision-Language-Action Models with Action Caching and Refinement

T0 review · 3 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read ActionCache claims that reusing past robot actions from a lightweight external cache can replace most of the iterative denoising in flow-based VLA models, accelerating action-head inference by up to 40x without any retraining, while preserv

desk verdict A solid, honest systems paper on training-free action caching for flow-based VLAs; the core claim holds up, but the threshold tuning and the untested same-scene/different-instruction regime need scrutiny. read the letter →

arxiv 2607.06370 v2 pith:3DHMPXNN submitted 2026-07-07 cs.RO cs.CVcs.LG

Ryuji Oi , Hikari Otsuka , Kosuke Matsushima , Yuki Ichikawa , Masato Motomura , Tatsuya Kaneko , Daichi Fujiki This is my paper · ORCID
classification cs.ROcs.CVcs.LG
keywords vision-language-actionmodelsflowmatchingactioncachingtraining-freeaccelerationwarm-startinferencerandomprojectionrobotmanipulationlatencyreduction
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

The paper argues that the expensive multi-step denoising used by flow-based vision-language-action (VLA) models is largely redundant across episodes and tasks, because robotic control repeatedly encounters similar visual and linguistic contexts and similar contexts call for similar action trajectories. ActionCache stores previously generated action chunks, each paired with a compact key derived from the model's own output embeddings via a fixed random projection, and retrieves the nearest past action for new contexts. When the retrieved context is similar enough, the cached action is used directly or with a single refinement step, skipping most of the action head's computation. In experiments, this yields action-head speedups of about 10x and 40x on two representative flow-based VLA models while keeping success rates within a few points of the full model. The central claim is that output-level retrieval can generalize warm-starting beyond local temporal continuity, turning past compute into reusable memory without modifying the pretrained policy.

What carries the argument

The central mechanism is the retrieve-and-warm-start cache. Each entry stores an intermediate noisy action chunk taken at a fixed denoising step of the full generation, paired with a cache key: a 500-dimensional sparse random projection (a fixed ternary matrix with no learning) of the VLM's output embeddings, augmented with robot-state features where the VLM lacks them. At inference, cosine similarity between the current key and stored keys triggers a hit or miss; a hit initializes zero- or few-step flow generation, while a miss runs the full denoising schedule. Cache management (LFU/LRU replacement, a pending buffer committed only on successful episodes, and a conservative hit threshold) ma

What would settle it

A controlled experiment where two visually near-identical scenes with substantially different language instructions produce high top-1 key similarity but require different actions: for instance, in the paper's push-button setup, compare cache-hit success on episodes where the retrieved entry came from a different instruction than the current one; if those hits fail disproportionately often, the key-similarity-to-action-similarity link is unsound.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the conditional flow-matching trajectory from noise to a valid action chunk is highly reusable: an action chunk generated in a past successful episode, retrieved by cosine similarity of sparse random projections of VLM output embeddings, sits close enough to the current target action that the model can either execute it directly at zero function evaluations or refine it in one step, with success rates comparable to the full multi-step generation. ActionCache operationalizes this as a plug-and-play external cache, committing only successful episodes and falling back to full generation when the similarity threshold is not met.

Load-bearing premise

Cosine similarity between fixed random projections of the model's output embeddings is a reliable enough proxy for the similarity of the actions the flow head would generate, so a close key-space match guarantees a close action-space match.

Editorial extensions

If this is right

  • A pretrained flow-based VLA can be accelerated 10.44x and 40.17x at the action head for two representative models while keeping success rates within a few points of the full-step model.
  • Action reuse generalizes across tasks: actions cached on one task support meaningful progress on different unseen tasks, particularly in early reaching phases, with fallback preserving the original policy's behavior.
  • ActionCache combines cleanly with existing VLM-level caching, yielding end-to-end latency reduction without degrading success rate, showing it is an orthogonal acceleration axis.
  • Cache size and hit threshold provide explicit latency-accuracy trade-off knobs, and performance remains stable under long-span deployment with continuous cache replacement.

Reading between the lines

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

  • The load-bearing link between embedding-key similarity and action similarity is likely to weaken at task boundaries where visually similar scenes require quite different actions (e.g., the same object layout but different language instructions); the authors' own pick_sausage results, where distractor objects cut the hit rate to 41.8%, suggest the margin is not large and would degrade further in le
  • If the redundancy the paper exploits is a general property of flow-based VLA conditioning, then caching could serve as a reusable non-parametric policy component shared across robots or deployments of the same model, a direction the authors did not explore.
  • A testable extension would replace the fixed random projection with a learned or session-adaptive key that explicitly includes instruction identity or camera-relative state; the paper's threshold sensitivity analysis indicates this could extend reuse to harder manipulation tasks.
  • The near-100% cache hit rates reported in some settings imply that the tested benchmarks are highly repetitive; a natural next experiment is to measure hit-rate and success-rate degradation as task diversity increases, which would bound the method's real-world applicability.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. ActionCache is a training-free external memoization layer for flow-matching action heads of VLA models. It computes a compact cache key as a sparse random projection of VLM output embeddings (Eq. 2), queries a cache of intermediate action chunks from successful episodes, and on cosine-similarity above a threshold Thit either executes the retrieved chunk directly (NFE=0) or uses it as initialization for N_hit refinement steps; otherwise it falls back to full generation. The paper evaluates on VLABench (10 primitive tasks) and LIBERO with π0.5 and GR00T-N1.6, and on three real-robot tasks, reporting up to 10.44× and 40.17× action-head speedups at roughly preserved success rates. It also presents cross-task reuse experiments, cache replacement policy comparison, hyperparameter sensitivity, long-span deployment behavior, and combination with VLA-Cache.

Significance. The central results are plausible and practically useful: in the evaluated repetitive-task regime, the reported numbers (Table 1, Table 2, Table 7) support the claim that success rates are preserved while action-head latency drops substantially. Strengths include disjoint prefill/test seeds, real-world validation, a conservative miss fallback, and a reasonably thorough hyperparameter sensitivity analysis (Appendix C). The cross-task claim is less strongly supported than the abstract suggests, and the main comparison table is malformed. If the robustness and presentation issues are addressed, ActionCache would be a solid contribution to low-latency VLA deployment.

major comments (3)
  1. [Section 4.2, Table 1] Table 1 is garbled and cannot serve as the primary evidence for the headline claim. In the π0.5 half, the base-model row below NFE=10 reads "16.8 ±2.9 2.5 7.52×" with no NFE label; the EfficientVLA rows read "217.95.2 3.62×" and "17.33.6 5.22×"; the GR00T-N1.6 rows are similarly truncated. A reader cannot tell which success rate corresponds to which NFE, or verify the claim that "base models drop their success rates severely at NFE=1". Please regenerate the table with explicit NFE values and separated columns, and include standard deviations for all methods.
  2. [Section 4.3 / Abstract] The "different tasks" claim in the abstract and contributions is not established by the cross-task experiment. Section 4.3 pre-fills with select_fruit and evaluates only select_painting and select_toy — three instances of the same select-X task family, with nearly identical language templates and workspace. It reports progress scores rather than success rates and uses an unusually strict threshold (Thit=0.925), so it cannot test the failure mode where high key similarity arises from shared visual/instructional context but the required action differs (e.g., same scene, different target object or instruction). Since NFE=0 executes a retrieved chunk directly, this regime is safety-critical. Please add an explicit same-scene/different-goal experiment (or a synthetic ablation swapping instructions) and report success rates, or qualify the cross-task claim in the abstract and conclusion.
  3. [Section 4.1 and Section 5 (Limitations)] The acceptance threshold Thit is a free parameter selected per task from the prefill similarity distribution ("shoulder of the hit-rate curve"; values 0.85, 0.875, 0.925 are used in different experiments), and it directly controls the hit/miss boundary. Figure 4c shows the cache hit rate swings from ~100% to ~0% over the Thit range, so the reported latency–success trade-off is partly determined by this subjective choice. The Limitations section acknowledges the extra hyperparameters but does not give an operational rule for setting Thit on a new task. Please specify a reproducible selection procedure (e.g., a fixed quantile of the prefill nearest-neighbor distribution) or report sensitivity of the headline numbers to Thit for each benchmark.
minor comments (5)
  1. [Section 4.2] EfficientVLA and Falcon are evaluated on 200 episodes with a single noise seed, while ActionCache and base models use 800 episodes with four seeds. Report standard deviations or confidence intervals for all baselines.
  2. [Section 3.3] Clarify the indexing of the stored intermediate action for the boundary cases N_hit=0 and N_hit=N. The phrase "denoising step N − N_hit" is ambiguous when N_hit=0.
  3. [Table 3] Table 3 is also poorly formatted: the VLA-Cache-only row shows concatenated latency values, and the ActionCache+VLA-Cache row reports overall latency 57.0 ms versus ActionCache-only 45.9 ms. Explain why combining with VLA-Cache increases overall latency, or fix the table.
  4. [Figure 3] The horizontal axes are inconsistent across subplots (some show 0–10, others 0–4). Please unify the axis ranges and labels.
  5. [Section 4.1 / Section 4.2] The default cache size is 3,000 in Section 4.1, but Section 4.2 uses 10,000 (π0.5) and 30,000 (GR00T-N1.6). State the cache size used for each table/figure and justify the discrepancy.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: core results are held-out success-rate measurements against external base models, with no load-bearing self-citation or definitional identity.

full rationale

ActionCache is an empirical systems paper with no derivation chain that reduces to its inputs. The method stores actions from successful episodes and retrieves them via cosine similarity of projected VLM embeddings; the paper does not define action similarity in terms of key similarity, nor does it fit a parameter to the success labels it later reports. The central VLABench and LIBERO results use disjoint prefill and test seeds: 'Prefill and test episodes use disjoint random seeds, ensuring no cached action is generated under an identical task configuration (e.g., object placement) to any test episode' (Sec. 4.1). Thus the reported success rates are held-out measurements, not reconstructions of the cache content. The hit threshold T_hit is selected by 'profiling the similarity distribution of the prefill phase and identifying the shoulder of the hit-rate curve before it plateaus' (Sec. 4.1); this is standard hyperparameter selection based on prefill statistics, not a fitted parameter that forces the reported success rates. The key-similarity-to-action-similarity link is an empirical design heuristic, not a definitional identity; it is tested indirectly via success rates and top-1 similarity (Fig. 4a) and is a legitimate correctness/robustness concern (e.g., same-scene/different-goal regimes), but that is not circularity. There are no self-citations: the reference list shows no overlap with the present authors, and external citations (random projections, LRU/LFU, prior warm-starting and caching methods) are used as standard background, not as an unverified self-consistency chain. The real-world push_button task may contain near-exact cached configurations because of its small discrete state space, but the paper's broader simulation results with disjoint seeds provide independent support for the cross-episode reuse claim. Overall, the derivation is self-contained against external benchmarks and no circular step was identified.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The method contributes an engineering mechanism, not a derivation; its load-bearing premises are empirical. The free-parameter count is high (six-plus deployment knobs plus one random projection draw), but they are disclosed in Table 5 and their sensitivity is partially charted in Appendix C. The most consequential premise — embedding-key similarity implies action similarity — is validated only on repetitive primitive tasks, which is the main external-validity risk. No new physical entities are postulated.

free parameters (7)
  • hit threshold Thit = 0.85 (π0.5 default), 0.65 (GR00T), 0.875/0.85 real-world, 0.925 cross-task, 0.88 long-span
    Acceptance threshold for cache hits; selected by profiling the prefill similarity distribution on the same benchmark (Sec. 4.1) and per task in the real world 'to maintain base model's success rate' (B.4).
  • refinement steps Nhit = 0 / 1 / 2 evaluated per experiment
    Number of flow-matching steps on a hit; directly trades latency vs success (Table 1). The 'maintains SR' claim holds at NFE=1-2, not uniformly at NFE=0 (LIBERO Long drops to 83.4).
  • cache capacity = 3,000 default; 10,000 (π0.5) / 30,000 (GR00T) in Table 1; 10,000 LIBERO; 300/1,000 real-world
    Appendix C shows cache size is 'the most important factor' for success rate; capacity is chosen per experiment, with no principled scaling rule.
  • cache-key dimension d = 500
    Projected key dimension; authors find d=300 sufficient (Appendix C) and choose 500 as 'sufficiently large' — a hand-picked value, though shown insensitive.
  • projection non-zero density p = 0.01
    Sparse ternary random projection density; insensitive within the evaluated range (Appendix C).
  • random projection matrix draw = seed not pinned in the paper (figures use seeds 42 and 44)
    R is 'created once and kept fixed' (Sec. 3.2); with d=500 over D≈2M the draw could in principle matter, though the seed-sensitivity plot (Fig. 8) suggests it does not.
  • cache key source = VLM output embeddings; VLM output + robot-state for GR00T
    Design choice ablated in Table 4, where VLM-out beats VLM-in by up to 29.1% absolute SR under similar hit rates.
assumptions (5)
  • standard math Very sparse random projections preserve cosine similarity well enough for nearest-neighbor retrieval at d=500 (Johnson-Lindenstrauss-type behavior).
    Invoked in Sec. 3.2 Eq. 2 without proof; standard result, but practical fidelity at d=500 over D≈2M is empirical and only mildly supported by the d-sensitivity ablation (Fig. 8).
  • domain assumption Cosine similarity in projected VLM output embedding space is a reliable proxy for similarity of the action chunks the flow head would generate (key-similarity → action-similarity).
    The load-bearing retrieval premise (Sec. 3.2-3.3). Supported only empirically on 10 repetitive VLABench tasks and 3 real-world tasks; distractor sensitivity in pick_sausage shows the premise is fragile (Sec. 4.6).
  • domain assumption A cached intermediate noisy action from a past successful generation is a valid initialization for the current conditional flow ODE, so that Nhit velocity-field evaluations correct retrieval error to the target action.
    The warm-start mechanism shared with prior diffusion-policy warm-start literature (Sec. 1 and Eq. 1); applied here across episodes/tasks rather than adjacent timesteps, a stronger assumption validated only through success-rate measurements.
  • domain assumption Disjoint prefill/test seeds guarantee no evaluation episode shares a task configuration with any cached configuration.
    Sec. 4.1/B.3: the task-setting seed and episode ID 'uniquely determine' the task configuration, ensuring no exact match. The guarantee is only as strong as the benchmark's seed→configuration entropy; on 10 primitive tasks with limited pose variation, near-duplicate configurations remain likely, so NFE=0 results may partly reflect reuse of near-identical successful trajectories.
  • ad hoc to paper Only successful episodes populate the cache (success-filtered population), biasing the cache toward high-quality actions.
    Sec. 3.3: pending entries are committed to the main cache only if the episode completes successfully. This design choice underlies the NFE=0 retrieval-policy behavior and is not present in prior diffusion caching.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ActionCache: Training-Free Acceleration for Vision-Language-Action Models with Action Caching and Refinement." pith.science (2026). https://pith.science/paper/3DHMPXNN

@misc{pith2026260706370,
  author       = {Pith},
  title        = {Pith review of: ActionCache: Training-Free Acceleration for Vision-Language-Action Models with Action Caching and Refinement},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3DHMPXNN}},
  note         = {Machine review of arXiv:2607.06370}
}
abstract

Vision-Language-Action (VLA) models have emerged as a promising approach for generalizable robotic manipulations. In particular, flow-matching-based VLA models have shown remarkable success due to their capability to generate precise and smooth action sequences and capture multimodal distributions. However, the iterative denoising process in the action head acts as a major computational bottleneck, posing a critical challenge for real-time deployment. To address this challenge, we propose ActionCache, a plug-and-play external cache that opportunistically reuses past intermediate actions to warm-start generations from the vicinity of target actions, drastically reducing the inference latency. Specifically, ActionCache stores the intermediate actions with compact multimodal keys, which enables retrieval from similar past contexts across different episodes or even different tasks. Experimental results in simulation and real-world environments demonstrate that ActionCache maintains high task success rates in a low-latency regime, achieving action head inference acceleration of up to $10.44\times$ and $40.17\times$ for representative flow-based VLA, $\pi_{0.5}$ and GR00T-N1.6, respectively.

Figures

Figures reproduced from arXiv: 2607.06370 by the authors.

Figure 1
Figure 1. Inference latency break￾down for representative VLA models on NVIDIA RTX 4090 GPU. Action head accounts for 37- 66% of the end-to-end latency. As shown in Equation (1), generating each action chunk in a flow-based VLA requires multi-step flow integration, where the action head repeatedly evaluates the velocity field Vθ. Thus, the number of flow matching steps directly determines the number of action head forward pas… view at source ↗
Figure 1
Figure 1. Top: Inference latency breakdown on NVIDIA RTX 4090 GPU for representative VLA models. Action head accountsfor 37-66% of the end-to-end latency. Bottom: ActionCache overview. ActionCache reduces the inference latency of action head by reusing past actions. and 39% for GR00T-N1.6 (Bjorck et al. 2025), as shown in [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of ActionCache. ActionCache is an external cache that stores intermediate noisy actions from past generations. When a cache hit occurs, ActionCache initializes from retrieved past similar actions, thereby reducing the number of denoising steps. 3.1 Framework Overview [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figures from the paper (16 more)
Figure 2
Figure 2. Figure 2: ActionCache Framework. ActionCache is an external cache that stores intermediate noisy actions from past genera￾tions. On a cache hit, ActionCache initializes from retrieved past similar actions, reducing the number of denoising steps. reduces the iteration cost of act…
Figure 3
Figure 3. Figure 3: Success rate and latency trade-off for ActionCache and baselines. (a): ActionCache achieves higher success rates throughout all NFEs compared to the baselines. (b): As the Nhit decreases, the latency of the action head decreases proportionally with a negligible latency…
Figure 3
Figure 3. Figure 3: Success rate and average NFEs on VLABench. ActionCache provides a better trade-off of NFE and success rate. tasks are executed in a round-robin manner, which means the cache contains entries from all the tasks. The latency is mea￾sured on a single NVIDIA RTX 5090 GPU w…
Figure 4
Figure 4. Figure 4: Cache hit rate across episode timesteps. Cache hit rate reaches over 80% at episode start stage. To investigate the potential for cross-task action reuse, we conduct a zero-shot cross-task experiment on VLABench with π0.5. In this experiment, we first populate the cach…
Figure 4
Figure 4. Figure 4: (a): Average Top-1 similarity vs success rate for various numbers of denoising steps. As the Top-1 similarity increases, the success rates improve. (b): Average Top1 similarity vs hit thresholds for various cache sizes. As the hit threshold increases, average Top-1 sim…
Figure 5
Figure 5. Figure 5: (a): Average Top-1 similarity vs success rate for various numbers of denoising steps. As the Top-1 similarity increases, the success rates improve. (b): Average Top1 similarity vs hit thresholds for various cache size. As the hit threshold increases, average Top-1 simi…
Figure 5
Figure 5. Figure 5: Progress score on the target tasks and cache hit rate across rollout timesteps. ActionCache maintains the base model’s progress scores in unseen tasks. Cache hit rate reaches over 80% at the starting stage. 4.3 Cross-Task Action Reuse To investigate the potential for c…
Figure 6
Figure 6. Figure 6: Hit rate comparison for cache replacement policies un￾der a restricted cache size of 100. LFU policy demonstrates the high￾est hit rate across hit thresholds. Comparison of Cache Replacement Policies under Re￾stricted Capacity. As established in Section 4.4, the qualit…
Figure 6
Figure 6. Figure 6: Hit rate comparison for cache replacement poli￾cies under a restricted cache size of 100. LFU policy demonstrates the highest hit rate across hit thresholds. Comparison of Cache Replacement Policies. As estab￾lished in Section 4.4, the quality of retrieved actions is i…
Figure 7
Figure 7. Figure 7: Real-world trajectories and corresponding cache hit rates. The cache hit rate is high on average during the approaching and placing phases, while it drops during the grasping phase. The video can be found in the supplementary material. specific and difficult grasps, LF…
Figure 7
Figure 7. Figure 7: Real-world Experimental Environment. We use 6-DoF SO-101 robotic arm with an overhead and wrist-mounted camera. B.2 Hyperparameter Settings [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Real-world experimental environment. We use 6-DoF SO-101 robotic arm with an overhead and wrist-mounted camera, a box and a white cube. B Results for LIBERO benchmark LIBERO consists of four task suites (Spatial, Object, Goal and Long) which contain 10 tasks each. We e…
Figure 8
Figure 8. Figure 8: Task success rates for hyperparameter changes. Only one parameter varies and the others are fixed. Cache key dimension d and cache size mainly affect the success rate, while the others do not. 42 44 Projection Matrix Seed 97 98 99 100 Cache Hit Rate [%] 10 2 10 3 Cache…
Figure 9
Figure 9. Figure 9: Cache hit rates for hyperparameter changes. Only one parameter varies and the others are fixed. The hit threshold Thit mainly affects the cache hit rate [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Moving average of task success rate and cache hit rate. The red and green vertical line shows the timings when the cache becomes full and when the moving average no longer contains prefill episodes. D Long-span Caching Evaluation Once the cache becomes full, ActionCac…
Figure 11
Figure 11. Figure 11: Visualization of Top-1 similarity and task phase for the three manipulation tasks. [PITH_FULL_IMAGE:figures/full_fig_p014_11.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. PhyAI: Real-Time Physical AI at the Edge, Scalable Rollouts in the Cloud

    cs.AI 2026-08 conditional novelty 6.0 of 10

    PhyAI unifies VLA and world-action model inference across onboard, edge, and cloud deployments with model adapters, reporting 1.40x to 4.65x latency gains over official baselines.

Reference graph

Works this paper leans on

53 extracted references · 8 linked inside Pith · cited by 1 Pith paper

  1. [1]

    Efficient

    Yantai Yang and Yuhao Wang and Zichen Wen and Luo Zhongwei and Chang Zou and Zhipeng Zhang and Chuan Wen and Linfeng Zhang , booktitle=. Efficient

  2. [2]

    Siyu Xu and Yunke Wang and Chenghao Xia and Dihao Zhu and Tao Huang and Chang Xu , booktitle=

  3. [3]

    2025 , booktitle=

    Qinglun Zhang and Zhen Liu and Haoqiang Fan and Guanghui Liu and Bing Zeng and Shuaicheng Liu , title=. 2025 , booktitle=

    2025

  4. [4]

    2026 , eprint=

    SnapFlow: One-Step Action Generation for Flow-Matching VLAs via Progressive Self-Distillation , author=. 2026 , eprint=

    2026

  5. [5]

    2506.12723 , archivePrefix=

    Ye Li and Yuan Meng and Zewen Sun and Kangye Ji and Chen Tang and Jiajun Fan and Xinzhu Ma and Shutao Xia and Zhi Wang and Wenwu Zhu , year=. 2506.12723 , archivePrefix=

    arXiv

  6. [6]

    2509.05614 , archivePrefix=

    Hanzhen Wang and Jiaming Xu and Jiayi Pan and Yongkang Zhou and Guohao Dai , year=. 2509.05614 , archivePrefix=

    arXiv

  7. [7]

    2506.13725 , archivePrefix=

    Wenxuan Song and Jiayi Chen and Pengxiang Ding and Yuxin Huang and Han Zhao and Donglin Wang and Haoang Li , year=. 2506.13725 , archivePrefix=

    arXiv

  8. [8]

    2503.20384 , archivePrefix=

    Rongyu Zhang and Menghang Dong and Yuan Zhang and Liang Heng and Xiaowei Chi and Gaole Dai and Li Du and Yuan Du and Shanghang Zhang , year=. 2503.20384 , archivePrefix=

    arXiv

Show all 53 references
  1. [9]

    Guanxing Lu and Zifeng Gao and Tianxing Chen and Wenxun Dai and Ziwei Wang and Yansong Tang , journal=

  2. [10]

    Yang Yue and Yulin Wang and Bingyi Kang and Yizeng Han and Shenzhi Wang and Shiji Song and Jiashi Feng and Gao Huang , booktitle=. DeeR-

  3. [11]

    Kevin Black and Noah Brown and James Darpinian and Karan Dhabalia and Danny Driess and Adnan Esmail and Michael Robert Equi and Chelsea Finn and Niccolo Fusai and Manuel Y. Galliker and Dibya Ghosh and Lachy Groom and Karol Hausman and brian ichter and Szymon Jakubczak and Tim...

  4. [12]

    2026 , eprint=

    _0 : A Vision-Language-Action Flow Model for General Robot Control , author=. 2026 , eprint=

    2026

  5. [13]

    2506.01844 , archivePrefix=

    Mustafa Shukor and Dana Aubakirova and Francesco Capuano and Pepijn Kooijmans and Steven Palma and Adil Zouitine and Michel Aractingi and Caroline Pascal and Martino Russi and Andres Marafioti and Simon Alibert and Matthieu Cord and Thomas Wolf and Remi Cadene , year=. 2506.01...

    arXiv

  6. [14]

    Moo Jin Kim and Karl Pertsch and Siddharth Karamcheti and Ted Xiao and Ashwin Balakrishna and Suraj Nair and Rafael Rafailov and Ethan P Foster and Pannag R Sanketi and Quan Vuong and Thomas Kollar and Benjamin Burchfiel and Russ Tedrake and Dorsa Sadigh and Sergey Levine and ...

  7. [15]

    Brianna Zitkovich and Tianhe Yu and Sichun Xu and Peng Xu and Ted Xiao and Fei Xia and Jialin Wu and Paul Wohlhart and Stefan Welker and Ayzaan Wahid and Quan Vuong and Vincent Vanhoucke and Huong Tran and Radu Soricut and Anikait Singh and Jaspiar Singh and Pierre Sermanet an...

  8. [16]

    Wenyao Zhang and Hongsi Liu and Zekun Qi and Yunnan Wang and XinQiang Yu and Jiazhao Zhang and Runpei Dong and Jiawei He and He Wang and Zhizheng Zhang and Li Yi and Wenjun Zeng and Xin Jin , booktitle=. Dream

  9. [17]

    Robotics: Science and Systems , year=

    Diffusion policy: Visuomotor policy learning via action diffusion , author=. Robotics: Science and Systems , year=

  10. [18]

    Wen, Junjie and Zhu, Yichen and Li, Jinming and Zhu, Minjie and Tang, Zhibin and Wu, Kun and Xu, Zhiyuan and Liu, Ning and Cheng, Ran and Shen, Chaomin and Peng, Yaxin and Feng, Feifei and Tang, Jian , journal=

  11. [19]

    2502.05855 , archivePrefix=

    Junjie Wen and Yichen Zhu and Jinming Li and Zhibin Tang and Chaomin Shen and Feifei Feng , year=. 2502.05855 , archivePrefix=

    arXiv

  12. [20]

    2411.19650 , archivePrefix=

    Qixiu Li and Yaobo Liang and Zeyu Wang and Lin Luo and Xi Chen and Mozheng Liao and Fangyun Wei and Yu Deng and Sicheng Xu and Yizhong Zhang and Xiaofan Wang and Bei Liu and Jianlong Fu and Jianmin Bao and Dong Chen and Yuanchun Shi and Jiaolong Yang and Baining Guo , year=. 2...

    arXiv

  13. [21]

    Goyal, Ankit and Hadfield, Hugo and Yang, Xuning and Blukis, Valts and Ramos, Fabio , journal=

  14. [22]

    Joshi and Ryan Julian and Dmitry Kalashnikov and Yuheng Kuang and Isabel Leal and Kuang

    Anthony Brohan and Noah Brown and Justice Carbajal and Yevgen Chebotar and Joseph Dabis and Chelsea Finn and Keerthana Gopalakrishnan and Karol Hausman and Alexander Herzog and Jasmine Hsu and Julian Ibarz and Brian Ichter and Alex Irpan and Tomas Jackson and Sally Jesmonth an...

  15. [23]

    International Conference on Learning Representations , year=

    Vision-language foundation models as effective robot imitators , author=. International Conference on Learning Representations , year=

  16. [24]

    Bo Liu and Yifeng Zhu and Chongkai Gao and Yihao Feng and qiang liu and Yuke Zhu and Peter Stone , booktitle=

  17. [25]

    2412.18194 , archivePrefix=

    Shiduo Zhang and Zhe Xu and Peiju Liu and Xiaopeng Yu and Yuan Li and Qinghui Gao and Zhaoye Fei and Zhangyue Yin and Zuxuan Wu and Yu-Gang Jiang and Xipeng Qiu , year=. 2412.18194 , archivePrefix=

    arXiv

  18. [26]

    2025 , journal=

    Johan Bjorck and Fernando Castañeda and Nikita Cherniadev and Xingye Da and Runyu Ding and Linxi and Yu Fang and Dieter Fox and Fengyuan Hu and Spencer Huang and Joel Jang and Zhenyu Jiang and Jan Kautz and Kaushil Kundalia and Lawrence Lao and Zhiqi Li and Zongyu Lin and Kevi...

    2025

  19. [27]

    2024 , booktitle =

    Agarwal, Shubham and Mitra, Subrata and Chakraborty, Sarthak and Karanam, Srikrishna and Mukherjee, Koyel and Saini, Shiv Kumar , title =. 2024 , booktitle =

    2024

  20. [28]

    Computer Vision and Pattern Recognition , year =

    Ma, Xinyin and Fang, Gongfan and Wang, Xinchao , title =. Computer Vision and Pattern Recognition , year =

  21. [29]

    2025 , booktitle =

    Xia, Yuchen and Sharma, Divyam and Yuan, Yichao and Kundu, Souvik and Talati, Nishil , title =. 2025 , booktitle =

    2025

  22. [30]

    2602.08245 , archivePrefix=

    Jinhao Li and Yuxuan Cong and Yingqiao Wang and Hao Xia and Shan Huang and Yijia Zhang and Ningyi Xu and Guohao Dai , year=. 2602.08245 , archivePrefix=

    arXiv

  23. [31]

    Real-time iteration scheme for diffusion policy

    Yufei Duan and Hang Yin and Danica Kragic. Real-time iteration scheme for diffusion policy. International Conference on Intelligent Robots and Systems. 2025

    2025

  24. [32]

    International Conference on Machine Learning , year=

    Falcon: Fast visuomotor policies via partial denoising , author=. International Conference on Machine Learning , year=

  25. [33]

    International Conference on Robotics and Automation , pages=

    Fast Policy Synthesis with Variable Noise Diffusion Models , author=. International Conference on Robotics and Automation , pages=

  26. [34]

    Vision-language-action models for robotics: A review towards real-world applications , year=

    Kawaharazuka, Kento and Oh, Jihoon and Yamada, Jun and Posner, Ingmar and Zhu, Yuke , journal=. Vision-language-action models for robotics: A review towards real-world applications , year=

  27. [35]

    A survey on vision--language--action models for embodied

    Ma, Yueen and Song, Zixing and Zhuang, Yuzheng and Hao, Jianye and King, Irwin , journal=. A survey on vision--language--action models for embodied

  28. [36]

    2023 , journal =

    Yang, Ling and Zhang, Zhilong and Song, Yang and Hong, Shenda and Xu, Runsheng and Zhao, Yue and Zhang, Wentao and Cui, Bin and Yang, Ming-Hsuan , title =. 2023 , journal =

    2023

  29. [37]

    2023 , publisher =

    Song, Yang and Dhariwal, Prafulla and Chen, Mark and Sutskever, Ilya , title =. 2023 , publisher =

    2023

  30. [38]

    and Kaiser,

    Vaswani, Ashish and Shazeer, Noam and Parmar, Niki and Uszkoreit, Jakob and Jones, Llion and Gomez, Aidan N. and Kaiser,. Attention is all you need , year =

  31. [39]

    Vision-language models for vision tasks: A survey , year=

    Zhang, Jingyi and Huang, Jiaxing and Jin, Sheng and Lu, Shijian , journal=. Vision-language models for vision tasks: A survey , year=

  32. [40]

    CoRR , title =

    Ronneberger, Olaf and Fischer, Philipp and Brox, Thomas , ee =. CoRR , title =

  33. [41]

    Johnson, W. B. and Lindenstrauss, J. , title =. Contemporary Mathematics , year =

  34. [42]

    , title =

    Wang, Xin and Yu, Fisher and Dou, Zi-Yi and Darrell, Trevor and Gonzalez, Joseph E. , title =. 2018 , booktitle =

    2018

  35. [43]

    2017 , booktitle =

    Bolukbasi, Tolga and Wang, Joseph and Dekel, Ofer and Saligrama, Venkatesh , title =. 2017 , booktitle =

    2017

  36. [44]

    Empirical Methods in Natural Language Processing , year=

    Clipscore: A reference-free evaluation metric for image captioning , author=. Empirical Methods in Natural Language Processing , year=

  37. [45]

    International Conference on Learning Representations , year=

    Flow matching for generative modeling , author=. International Conference on Learning Representations , year=

  38. [46]

    International Conference on Learning Representations , year=

    Remi Cadene and Simon Alibert and Francesco Capuano and Michel Aractingi and Adil Zouitine and Pepijn Kooijmans and Jade Choghari and Martino Russi and Caroline Pascal and Steven Palma and Dana Aubakirova and Mustafa Shukor and Jess Moss and Alexander Soare and Quentin Lhoest ...

  39. [47]

    Belady, L. A. , title =. 1966 , journal =

    1966

  40. [48]

    Advances in Neural Information Processing Systems , year=

    Denoising diffusion probabilistic models , author=. Advances in Neural Information Processing Systems , year=

  41. [49]

    SIGKDD Conference on Knowledge Discovery and Data Mining , year =

    Very sparse random projections , author =. SIGKDD Conference on Knowledge Discovery and Data Mining , year =

  42. [50]

    James Bradbury and Roy Frostig and Peter Hawkins and Matthew James Johnson and Yash Katariya and Chris Leary and Dougal Maclaurin and George Necula and Adam Paszke and Jake Vander

  43. [51]

    PyTorch: an imperative style, high-performance deep learning library , year =

    Paszke, Adam and Gross, Sam and Massa, Francisco and Lerer, Adam and Bradbury, James and Chanan, Gregory and Killeen, Trevor and Lin, Zeming and Gimelshein, Natalia and Antiga, Luca and Desmaison, Alban and K\". PyTorch: an imperative style, high-performance deep learning libr...

  44. [52]

    Belady, L. A. , journal=. A study of replacement algorithms for a virtual-storage computer , year=

  45. [53]

    2026 , eprint=

    Let It Be Simple: One-Step Action Generation for Vision-Language-Action Models , author=. 2026 , eprint=

    2026

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.