Pith.

Robotics · arXiv:2611.21607 v1 · submitted 16 Nov 2026 · 113 references

ScaleVLA: Scaling Vision-Language-Action Models to 552 Billion Parameters

Ruixi Yu, Dairu Liu, Xuchuan Chen, Zekun Qi, Zekai Li, Zhuo Cao, Luyang Niu, Yingsheng Zhang

show 4 more authors

Sikai Liang, Ke Fan, He Wang, Li Yi

The paper says

“Frontier Vision-Language Models (VLMs) such as GPT-6 Astra have demonstrated remarkable capabilities in robot control, suggesting that broad pretraining provides knowledge valuable for physical action.”

more of the abstract

Yet most Vision-Language-Action (VLA) policies still rely on much smaller VLM backbones, leaving this potential largely untapped. We ask whether frontier-scale representations can provide a stronger foundation for robot learning. We introduce ScaleVLA, which connects a 552B DeepSeek-V4.1-Flash backbone to a trainable action DiT, and systematically compare VLMs from 2B to 552B under a shared training setup. We evaluate three fundamental capabilities of robot foundation models. In sample efficiency, just 10 demonstrations outperform a 2B backbone trained with 50 (87.8% vs. 85.6%). In generalization, LIBERO-PRO success reaches 45.1%, compared with at most 31.3% for smaller backbones. In precise manipulation, real-world charger insertion improves from 4/10 to 9/10 successes, with gains concentrated in contact-sensitive alignment. Despite its scale, ScaleVLA achieves 216 ms median latency per action chunk. Our findings suggest that frontier VLMs can serve not only as reasoning agents, but also as powerful representational foundations for efficient and generalizable robot policies.

from the abstract

Pith says

Machine review · deepseek-v4-flash · 17 Nov 2026 · read the full text (106,871 characters) · not checked by a person

It can be wrong. Challenge this review or reply to it.

Holds up, with conditions. ScaleVLA provides the first systematic comparison of VLM backbones up to 552B parameters for robot learning, showing clear gains in sample efficiency and generalization, but the headline claims are partly confounded by architecture type and token count. This line is wrong

8 minor

How often objections from reviews like this one are right: not yet measured

Are you an author? What you can do.

The paper, first character to last

The reviewer was given the whole text, 106,871 characters. Not checked: figures, code, what the 113 references say.

8 minor objections

The paper says, at §3.1

Pith says: minor objection 1 of 8

The description of the feature extraction layer selection (round(16L/28)) is unclear. Why 16/28? This should be justified or referenced.

The paper says, at §4.2, Table 5

Pith says: minor objection 2 of 8

The latency measurement uses vLLM for timing but notes that features are not compared element-wise with the Hugging Face deployment used for success rates. This discrepancy could affect latency numbers. Please clarify whether the timing setup is representative of actual inference.

show 6 more

The paper says, at §4.5

Pith says: minor objection 3 of 8

The real robot evaluation uses only one training seed per policy. While evaluation episodes are repeated, the lack of multiple training seeds limits statistical robustness.

The paper says, at §2

Pith says: minor objection 4 of 8

The related work section is comprehensive but could be more concise. Some references are not directly relevant to the scaling claim.

The paper says, at §4.1, Table 1

Pith says: minor objection 5 of 8

The table lists backbones but does not include the number of active parameters for the MoE model. This is important for understanding computational cost.

The paper says, at §4.2, Table 3

Pith says: minor objection 6 of 8

The LIBERO-PRO results show high variance across backbones (e.g., Qwen3-VL-32B drops to 22.8% mean). This suggests that some backbones may not be well-suited for the action head architecture. The paper should discuss why certain backbones perform poorly.

The paper says, at §4.3, Table 5

Pith says: minor objection 7 of 8

The latency table shows that the action head time is roughly constant (~70ms) across backbones. This is interesting but not discussed. It suggests that the action head is not the bottleneck for larger backbones.

The paper says, at §5

Pith says: minor objection 8 of 8

The conclusion states that frontier VLMs can serve as representational foundations, but the paper does not ablate the effect of freezing the backbone. Would fine-tuning the backbone improve performance further? This is a natural next step.

What would settle it

Re-run the LIBERO-Goal sample efficiency experiment (Table 6) with the 552B backbone but artificially truncating the input token count to match the 212 tokens used by smaller backbones (e.g., by subsampling image patches). If the performance at 10 demonstrations drops substantially (e.g., below 80%), then the token count difference is a significant confound. If performance remains similar, the backbone scale is the primary driver.

from the skeptic's reading

How the review went

5 passes by deepseek-v4-flash · 17 Nov 2026

  1. Reader holds up, with conditions confidence high
  2. Skeptic partly agrees 7 s · 36,392 tokens read, 682 written
  3. Referee 8 minor 12 s · 36,608 tokens read, 1,589 written
  4. Check against the paper kept 0 of 4 objections 13 s · 36,740 tokens read, 2,099 written
  5. Editor wrote the verdict above 6 s · 37,055 tokens read, 504 written

Not run on this paper: the simulated authors' defence.

Pith's passes raised 4 objections. A check against the paper's text kept 0 and withdrew 4 (3 the paper already says; 1 Pith's reviewer misread the paper).

The review in full

The review in brief, machine-written

This paper asks a straightforward question: if you keep the action head fixed and swap in larger VLM backbones, does robot policy performance improve? The answer, backed by experiments across LIBERO, RoboCasa, DexJoCo, and a real charger-insertion task, is yes—and the gains are often substantial. With 552B parameters (DeepSeek-V4.1-Flash, MoE), the policy learns from fewer demonstrations, generalizes better under distribution shift, and handles contact-rich manipulation more reliably than 2B–32B dense backbones. The 216 ms median latency per chunk, while requiring four H200 GPUs, shows that frontier-scale features can be practical for closed-loop control.

The paper does what scaling studies should do: it holds the action architecture, training data, and optimization budget fixed across backbones. The comparisons are extensive and the real-robot evaluation is a welcome addition. The central trend—larger backbones help—is credible and useful for the VLA community.

That said, the headline comparisons are not pure scaling studies. The 552B backbone uses a sparse MoE architecture, while all smaller backbones are dense. It also processes 568 input tokens versus 212 for the others. The paper acknowledges both confounds but does not control for them. The sample-efficiency claim (10 demos with 552B beats 50 demos with 2B) is based on a fixed training budget; the paper's own additional results show that smaller backbones catch up with more training. Most results lack error bars or multiple seeds, making it hard to assess significance. Code and data are not released.

These are addressable issues. The paper would be stronger with a controlled comparison that includes a dense model at a larger scale or an MoE model at a smaller scale, error bars, and released assets. As it stands, the broad conclusion is likely correct, but the specific quantitative claims should be read with caution.

This paper deserves peer review. It tackles a timely question with a serious empirical effort, and the confounds can be resolved in revision.

The strongest objection

The paper's central claim is that scaling the VLM backbone of a robot policy to 552B parameters substantially improves sample efficiency, generalization, and contact-rich manipulation compared to smaller backbones. The reader's weakest_assumption identifies a potential confound: the 552B backbone uses a sparse MoE architecture while smaller backbones are dense, and the 552B backbone processes more input tokens. However, the paper acknowledges these factors explicitly (lines [b150]–[b151]) and does not claim to have isolated them. The core argument is about the practical benefit of using a frontier-scale VLM as a representational foundation, not about isolating the effect of parameter count per se. The comparison is between available frontier-scale VLMs and available smaller VLMs, which is a realistic and useful comparison even if architecture and token count differ. The paper's own limitations section notes these confounds. The results are consistent across multiple benchmarks (LIBERO, LIBERO-PRO, RoboCasa, DexJoCo, real robot) and the gains are large and systematic. The sample efficiency claim is qualified by the paper's own additional results showing that smaller backbones catch up with more training (line [b261]). The lack of error bars is a concern, but the paper reports results from multiple evaluation episodes (50 per task for LIBERO, 500 for LIBERO-PRO, etc.) which provides some statistical grounding. The real-robot results are limited to a single task, but the simulation results cover diverse tasks. Overall, the central claim is well-supported by the evidence presented, and the acknowledged limitations do not undermine it.

The reader's verdict is CONDITIONAL, citing the confound between scale and architecture/token count as the weakest assumption. While this is a valid methodological concern, the paper acknowledges it and the central claim is about the practical benefit of using a frontier-scale VLM, not about isolating the effect of parameter count. The paper compares available models, which is a realistic evaluation. The results are consistent across multiple benchmarks and the gains are large. The lack of error bars is a concern but partially mitigated by the number of evaluation episodes. The sample efficiency claim is qualified by the paper's own additional results. The real-robot evaluation is limited but the simulation results are extensive. Overall, the evidence supports the central claim, and the acknowledged limitations do not warrant conditional acceptance. The paper would benefit from addressing the confound, but it is not necessary for the central claim to hold. I partially agree with the reader's identification of the weakest assumption, but I disagree that it is load-bearing enough to change the verdict.

The referee's report

Not published. Pith's referee wrote this assessment before its objections were checked against the paper, and 4 of its 4 major objections did not survive that check.

Pith's reading of the paper (a machine's summary, not the authors')

The paper investigates whether frontier-scale vision-language model (VLM) representations can serve as a stronger foundation for robot learning than the smaller backbones typically used in vision-language-action (VLA) policies. The authors introduce ScaleVLA, which connects a 552B-parameter DeepSeek-V4.1-Flash backbone to a trainable action diffusion transformer (DiT), and compare it against backbones from 2B to 32B parameters across simulation and real-world benchmarks. They report that the 552B backbone achieves higher success rates with fewer demonstrations, generalizes better under distribution shift, and performs more reliably on contact-rich manipulation tasks like charger insertion, while maintaining subsecond inference latency. The paper argues that frontier VLMs can serve not only as reasoning agents but also as powerful representational foundations for efficient and generalizable robot policies.

What carries the argument. The central object is ScaleVLA, which connects a 552B-parameter DeepSeek-V4.1-Flash VLM backbone (a sparse mixture-of-experts model) to a trainable action diffusion transformer (DiT). The VLM backbone encodes images and language instructions; the action DiT converts these representations into continuous robot actions via flow matching. The backbone is frozen, and only the action head (feature adapter, state/action encoders, DiT, action decoder) is trained on robot demonstrations. The key mechanism is that the large pretrained VLM provides rich representations that the action head can directly map to actions without autoregressive reasoning.

Core claim. The paper claims that scaling the VLM backbone of a robot policy to 552 billion parameters (using DeepSeek-V4.1-Flash, a sparse mixture-of-experts model) substantially improves sample efficiency, generalization under distribution shift, and contact-rich manipulation performance compared to smaller backbones (2B-32B), while maintaining subsecond inference latency. The specific quantitative claims include: on LIBERO-Goal, 10 demonstrations with 552B outperform 50 demonstrations with 2B (87.8% vs 85.6%); on LIBERO-PRO, 552B achieves 45.1% average success vs at most 31.3% for smaller backbones; on real charger insertion, 552B succeeds 9/10 vs 4/10 for 2B; median inference latency is 216 ms per action chunk on four H200 GPUs.

The premise it rests on. The paper assumes that the performance gains observed when scaling the VLM backbone from 2B to 552B are primarily due to the increased representational capacity of the larger backbone, and not confounded by other factors that differ between the backbones, such as architecture type (MoE vs dense) or input token count.

If the paper is right (Pith's inferences, not the authors')

If the claims hold, frontier VLMs could serve as a drop-in representational foundation for robot policies, potentially reducing the amount of robot demonstration data needed.

The results suggest that scaling VLM backbones may be a more effective path to improving robot policy generalization than scaling robot data alone.

The approach could enable more capable robot policies for contact-rich manipulation tasks that require precise spatial perception and control.

The subsecond inference latency (216 ms) suggests that frontier-scale VLMs can be used for practical real-time robot control, albeit with higher hardware requirements.

The paper does not control for architecture type (MoE vs dense) or input token count, so the observed gains may be partly due to these confounds rather than parameter count alone. A controlled comparison using a dense 552B model (if one existed) would be needed to isolate the effect of scale.

The paper's claim that frontier VLMs can serve as 'representational foundations' for robot policies suggests a potential shift in VLA research: instead of training or fine-tuning large VLMs on robot data, one could freeze a frontier VLM and train only a lightweight action head. This could make frontier-scale robot policies more practical.

The results on LIBERO-PRO (45.1% vs at most 31.3%) suggest that larger backbones may be particularly beneficial for out-of-distribution generalization, which is a key challenge for deploying robot policies in the real world.

Check this review

What was read

The passes were given all 106,871 characters of the paper's LaTeX source.

Fingerprint of that text (SHA-256): f974b3e0c71bffa42f1bb826ac13cb24a627b3e88f3fd18849c8eb5af95f37c5

Pith does not keep that text. Rebuilt from the paper's source or PDF by the same code, it has this fingerprint if it is the same text. From a copy of Pith's code:

PYTHONPATH=. python scripts/review_input_check.py 2611.21607

The passes, in the order they ran

  1. Reader deepseek-v4-flash · medium effort · $0.0048 · instructions of the Reader, fingerprint f9ead7300382…
  2. Skeptic deepseek-v4-flash · medium effort · $0.0034 · instructions of the Skeptic, fingerprint 8f88c0f61488…
  3. Referee deepseek-v4-flash · medium effort · $0.0036 · instructions of the Referee, fingerprint a852d99aea0d…
  4. Check against the paper deepseek-v4-flash · medium effort · $0.0037 · instructions of the Check against the paper, fingerprint 6c7a6883d551…
  5. Editor deepseek-v4-flash · medium effort · $0.0035 · instructions of the Editor, fingerprint e7211f6a3d27…
  6. Summary for the page deepseek-v4-flash · medium effort · temperature 0.2 · $0.0037 · instructions of the Summary for the page, fingerprint e4f56bf568e7…

The passes cost $0.023 in model calls together.

Version stamped on this review: pith-pipeline@v1.4.1-daily-deepseek-or-medium+deepseek/deepseek-v4-flash+p.0ef1a750a3e3

Each link opens a pass's instructions as plain text, the exact bytes its fingerprint was taken over. Every stored set of instructions.

curl -s https://pith.science/methodology/prompts/f9ead730038216984cfb56d7cb34332bdf691828296a1de4245f3d6667d4bf56 | sha256sum

prints the fingerprint of the reader's instructions. The same works for every pass.

The receipt

The review's fingerprint was taken on 17 Nov 2026. The review you are reading still matches it.

Fingerprint of the review (SHA-256): 85c11ee65b96fe3fcbe6d47609e9ed7a09ebccfae2eb0f3e3b4f149b8ebc965a

curl -s 'https://pith.science/api/v1/verdicts/2611.21607?form=canonical' | sha256sum

prints the fingerprint of the review as it stands now. The rule, in four steps.

A matching fingerprint shows the review has not changed since that date. It does not show that the review is right.

How often reviews like this one are right

Reviews like this one: written by deepseek-v4-flash, from the paper's LaTeX source.

How often objections from reviews like this one are right: not yet measured

Every value above, in the review's public record (JSON)

Reviews by people

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review

How Pith checks papers · The full audit of this paper

The paper

Abstract

Frontier Vision-Language Models (VLMs) such as GPT-6 Astra have demonstrated remarkable capabilities in robot control, suggesting that broad pretraining provides knowledge valuable for physical action. Yet most Vision-Language-Action (VLA) policies still rely on much smaller VLM backbones, leaving this potential largely untapped. We ask whether frontier-scale representations can provide a stronger foundation for robot learning. We introduce ScaleVLA, which connects a 552B DeepSeek-V4.1-Flash backbone to a trainable action DiT, and systematically compare VLMs from 2B to 552B under a shared training setup. We evaluate three fundamental capabilities of robot foundation models. In sample efficiency, just 10 demonstrations outperform a 2B backbone trained with 50 (87.8% vs. 85.6%). In generalization, LIBERO-PRO success reaches 45.1%, compared with at most 31.3% for smaller backbones. In precise manipulation, real-world charger insertion improves from 4/10 to 9/10 successes, with gains concentrated in contact-sensitive alignment. Despite its scale, ScaleVLA achieves 216 ms median latency per action chunk. Our findings suggest that frontier VLMs can serve not only as reasoning agents, but also as powerful representational foundations for efficient and generalizable robot policies.

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.

abstract compiled · 2026-11-17

The authors' main claim, as the classifier picked it out 1.00

“Our findings suggest that frontier VLMs can serve not only as reasoning agents, but also as powerful representational foundations for efficient and generalizable robot policies.”

How the main claim is stated: hedged possibility 0.99

What the abstract itself states, as probabilities: claims first 0.07 · states limitation 0.16 · human participants 0.06 · controlled comparison 0.93 · randomized assignment 0.02 · states quantitative result 0.99 · compares with prior results 0.96 · mentions code or data availability 0.03

The abstract, sentence by sentence

This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.

SentenceRoleThe authors' own workStates the main conclusionStrength
Frontier Vision-Language Models (VLMs) such as GPT-6 Astra have demonstrated remarkable capabilities in robot control, suggesting that broad pretraining provides knowledge valuable for physical action. background 1.00 0.05 0.04 hedged possibility 0.59
Yet most Vision-Language-Action (VLA) policies still rely on much smaller VLM backbones, leaving this potential largely untapped. gap 0.96 0.07 0.02 asserted 1.00
We ask whether frontier-scale representations can provide a stronger foundation for robot learning. gap 0.82 0.16 0.03 unclear 0.52
We introduce ScaleVLA, which connects a 552B DeepSeek-V4.1-Flash backbone to a trainable action DiT, and systematically compare VLMs from 2B to 552B under a shared training setup. contribution 0.70 0.96 0.05 asserted 0.98
We evaluate three fundamental capabilities of robot foundation models. method 0.59 0.80 0.02 asserted 0.68
In sample efficiency, just 10 demonstrations outperform a 2B backbone trained with 50 (87.8% vs. 85.6%). result 1.00 0.94 0.05 asserted 1.00
In generalization, LIBERO-PRO success reaches 45.1%, compared with at most 31.3% for smaller backbones. result 1.00 0.92 0.06 asserted 1.00
In precise manipulation, real-world charger insertion improves from 4/10 to 9/10 successes, with gains concentrated in contact-sensitive alignment. result 1.00 0.94 0.04 asserted 1.00
Despite its scale, ScaleVLA achieves 216 ms median latency per action chunk. result 1.00 0.94 0.03 asserted 1.00
Our findings suggest that frontier VLMs can serve not only as reasoning agents, but also as powerful representational foundations for efficient and generalizable robot policies. interpretation 1.00 0.76 0.92 hedged possibility 0.99

What kind of paper the classifier reports this is: method or tool 0.67 · evidence from observation or experiment 0.99

What was examined, and what was not

Sources read

  • arXiv listing abstract v1

Not examined

  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers

Levels run

  • P abstract

Not run

  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

How the labels were made

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-17: 1 request, 7,602 input tokens.

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON

References (113)
  1. [1] Shuai Bai, Yuxuan Cai, Ruizhe Chen, et al. Qwen3-VL technical report.arXiv preprint arXiv:2511.21631, 2025. https://arxiv.org/abs/ 2511.21631 on PithPith's reviewarXiv
  2. [2] Suneel Belkhale, Tianli Ding, Ted Xiao, Pierre Sermanet, Quan Vuong, Jonathan Tompson, Yev- gen Chebotar, Debidatta Dwibedi, and Dorsa Sadigh. RT-H: Action Hierarchies Using Language. InRobotics: Science and Systems, 2024. doi: 10.15607/RSS.2024.XX.049. https://doi.org/10. 156... on Pithdoi:10.15607/rss.2024.xx.049
  3. [3] Hongzhe Bi, Hengkai Tan, Shenghao Xie, Zeyuan Wang, Shuhe Huang, Haitian Liu, Ruowen Zhao, Yao Feng, Chendong Xiang, Yinze Rong, Hongyan Zhao, Hanyu Liu, Zhizhong Su, Lei Ma, Hang Su, and Jun Zhu. Motus: A Unified Latent Action World Model.arXiv preprint arXiv:2512.13030, 2025... on PithPith's reviewarXiv
  4. [4] Johan Bjorck, Fernando Castañeda, et al. GR00T N1: An open foundation model for generalist hu- manoid robots.arXiv preprint arXiv:2503.14734, 2025 on PithPith's reviewarXiv
  5. [5] Kevin Black, Noah Brown, Danny Driess, et al. π0: A vision-language-action flow model for general robot control.arXiv preprint arXiv:2410.24164, 2024 on PithPith's reviewarXiv
  6. [6] Konstantinos Bousmalis, Giulia Vezzani, Dushyant Rao, Coline Manon Devin, Alex X. Lee, Maria Bauzá Villalonga, Todor Davchev, Yuxiang Zhou, Agrim Gupta, Akhil Raju, Antoine Laurens, Claudio Fantacci, Valentin Dalibard, Martina Zam- belli, Murilo Fernandes Martins, Rugile Pevce... on Pith
  7. [7] Anthony Brohan, Noah Brown, Justice Carba- jal, Yevgen Chebotar, Joseph Dabis, Chelsea Finn, Keerthana Gopalakrishnan, Karol Hausman, Alexander Herzog, Jasmine Hsu, Julian Ibarz, Brian Ichter, Alex Irpan, Tomas Jackson, Sally Jesmonth, Nikhil J. Joshi, Ryan Julian, Dmitry Kala... on Pith
  8. [8] Qingwen Bu, Yanting Yang, Jisong Cai, Shenyuan Gao, Guanghui Ren, Maoqing Yao, Ping Luo, and Hongyang Li. UniVLA: Learning to Act Anywhere with Task-centric Latent Actions.arXiv preprint arXiv:2505.06111, 2025. https://arxiv.org/abs/ 2505.06111 on PithPith's reviewarXiv
  9. [9] Rahul Chand and Miguel Villa Floran. How many demonstrations does a VLA need? on the sam- ple efficiency of VLA finetuning.arXiv preprint arXiv:2611.09964, 2026. https://arxiv.org/abs/ 2611.09964 on PithPith's reviewarXiv
  10. [10] Chi-Lam Cheang, Guangzeng Chen, Ya Jing, Tao Kong, Hang Li, Yifeng Li, Yuxiao Liu, Hong- tao Wu, Jiafeng Xu, Yichu Yang, Hanbo Zhang, and Minzhao Zhu. GR-2: A Generative Video- 10 Language-Action Model with Web-Scale Knowl- edge for Robot Manipulation.arXiv preprint arXiv:2410... on PithPith's reviewarXiv
  11. [11] BoyuanChen, ZhuoXu, SeanKirmani, BrianIchter, Dorsa Sadigh, Leonidas J. Guibas, and Fei Xia. Spa- tialVLM: Endowing vision-language models with spatial reasoning capabilities. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14455–14465. IEEE, on Pith
  12. [12] Tianxing Chen, Yue Chen, Zixuan Li, Junyuan Tang, Kailun Su, Weijie Wan, Baijun Chen, Haoran Lu, Haowen Yan, Honghao Su, et al. RoboDojo: A unified sim-and-real benchmark for comprehensive evaluation of generalist robot manipulation policies. arXiv preprint arXiv:2607.04434, 2... on PithPith's reviewarXiv
  13. [13] Cheng Chi, Siyuan Feng, Yilun Du, Zhenjia Xu, Eric Cousineau, Benjamin Burchfiel, and Shuran Song. Diffusion policy: Visuomotor policy learn- ing via action diffusion. InRobotics: Science and Systems, 2023. doi: 10.15607/RSS.2023.XIX.026. https://doi.org/10.15607/RSS.2023.XIX.026 on Pithdoi:10.15607/rss.2023.xix.026
  14. [14] Damai Dai, Chengqi Deng, Chenggang Zhao, R.X. Xu, Huazuo Gao, Deli Chen, Jiashi Li, Wangding Zeng, Xingkai Yu, Y. Wu, Zhenda Xie, Y.K. Li, Panpan Huang, Fuli Luo, Chong Ruan, Zhifang Sui, and Wenfeng Liang. DeepSeekMoE: Towards ultimate expert specialization in mixture-of-expe... on Pith
  15. [15] DeepSeek-AI. DeepSeek-V4.1-Flash: Pushing the limits of KV cache compression. Technical report, on Pith
  16. [16] Ria Doshi, Homer Rich Walke, Oier Mees, Sudeep Dasari, and Sergey Levine. Scaling Cross-Embodied Learning: One Policy for Manipulation, Navi- gation, Locomotion and Aviation. InConfer- ence on Robot Learning, volume 270 ofProceed- ings of Machine Learning Research, pages 496–5... on Pith
  17. [17] doi: 10.18653/v1/2024.acl-long.70. https: //aclanthology.org/2024.acl-long.70/ on Pithdoi:10.18653/v1/2024.acl-long.70
  18. [18] Danny Driess, Jost Tobias Springenberg, Brian Ichter, LiliYu, AdrianLi-Bell, KarlPertsch, AllenZ. Ren, Homer Walke, Quan Vuong, Lucy Xiaoyang Shi, and Sergey Levine. Knowledge insulating vision-language-action models: Train fast, run fast, generalize better. InAdvances in Neur... on PithPith's reviewarXiv
  19. [19] Yilun Du, Sherry Yang, Bo Dai, Hanjun Dai, Ofir Nachum, Josh Tenenbaum, Dale Schuurmans, and Pieter Abbeel. Learning universal policies via text- guided video generation. InAdvances in Neural Information Processing Systems, volume 36, 2023. https://arxiv.org/abs/2302.00111 on PithPith's reviewarXiv
  20. [20] Galbot Team. Systematically exploring the capa- bilities of GPT-6 Astra as embodied policies, 2026. https://arxiv.org/abs/2609.38537 on PithPith's reviewarXiv
show 80 more
  1. [21] Danny Driess, Fei Xia, Mehdi S. M. Sajjadi, Corey Lynch, Aakanksha Chowdhery, Brian Ichter, Ayzaan Wahid, Jonathan Tompson, Quan Vuong, Tianhe Yu, Wenlong Huang, Yevgen Chebotar, Pierre Sermanet, Daniel Duckworth, Sergey Levine, Vincent Vanhoucke, Karol Hausman, Marc Tou- ssai... on Pith
  2. [22] Dibya Ghosh, Homer Rich Walke, Karl Pertsch, Kevin Black, Oier Mees, Sudeep Dasari, Joey Hejna, Tobias Kreiman, Charles Xu, Jianlan Luo, You Liang Tan, Lawrence Yunliang Chen, Quan Vuong, Ted Xiao, Pannag R. Sanketi, Dorsa Sadigh, Chelsea Finn, and Sergey Levine. Octo: An open... on Pithdoi:10.15607/rss.2024.xx.090
  3. [23] Ankit Goyal, Jie Xu, Yijie Guo, Valts Blukis, Yu- Wei Chao, and Dieter Fox. RVT: Robotic View Transformer for 3D Object Manipulation. InCon- ference on Robot Learning, volume 229 ofProceed- ings of Machine Learning Research, pages 694–710. PMLR, 2023. https://proceedings.mlr.p... on Pith
  4. [24] Ankit Goyal, Valts Blukis, Jie Xu, Yijie Guo, Yu-Wei Chao, and Dieter Fox. RVT-2: Learn- ing Precise Manipulation from Few Demonstra- tions. InRobotics: Science and Systems, 2024. doi: 10.15607/RSS.2024.XX.055. https://doi.org/10. 15607/RSS.2024.XX.055 on Pithdoi:10.15607/rss.2024.xx.055
  5. [25] Ryan Hoque, Peide Huang, David J. Yoon, Mouli Sivapurapu, and Jian Zhang. EgoDex: Learning Dexterous Manipulation from Large-Scale Egocen- tric Video.arXiv preprint arXiv:2505.11709, 2025. https://arxiv.org/abs/2505.11709. 11 on PithPith's reviewarXiv
  6. [26] Gemini Robotics Team. Gemini robotics: Bring- ing AI into the physical world.arXiv preprint arXiv:2503.20020, 2025 on PithPith's reviewarXiv
  7. [27] Wenlong Huang, Fei Xia, Ted Xiao, Harris Chan, Jacky Liang, Pete Florence, Andy Zeng, Jonathan Tompson, Igor Mordatch, Yevgen Chebotar, Pierre Sermanet, Tomas Jackson, Noah Brown, Linda Luu, Sergey Levine, Karol Hausman, and Brian Ichter. Inner Monologue: Embodied Reasoning th... on Pith
  8. [28] Wenlong Huang, Chen Wang, Ruohan Zhang, Yunzhu Li, Jiajun Wu, and Li Fei-Fei. Vox- Poser: Composable 3D value maps for robotic manipulation with language models. InConfer- ence on Robot Learning, volume 229 ofProceed- ings of Machine Learning Research, pages 540–562. PMLR, 202... on Pith
  9. [29] Wenlong Huang, Chen Wang, Yunzhu Li, Ruo- han Zhang, and Li Fei-Fei. ReKep: Spatio- Temporal Reasoning of Relational Keypoint Con- straints for Robotic Manipulation. InConfer- ence on Robot Learning, volume 270 ofProceedings of Machine Learning Research, pages 4573–4602. PMLR,... on Pith
  10. [30] Brian Ichter, Anthony Brohan, Yevgen Chebotar, Chelsea Finn, Karol Hausman, Alexander Herzog, Daniel Ho, Julian Ibarz, Alex Irpan, Eric Jang, Ryan Julian, Dmitry Kalashnikov, Sergey Levine, Yao Lu, Carolina Parada, Kanishka Rao, Pierre Sermanet, Alexander Toshev, Vincent Vanho... on Pith
  11. [31] Siyuan Huang, Zhengkai Jiang, Hao Dong, Yu Qiao, Peng Gao, and Hongsheng Li. Instruct2Act: Map- ping Multi-modality Instructions to Robotic Ac- tions with Large Language Model.arXiv preprint arXiv:2305.11176, 2023. https://arxiv.org/abs/ 2305.11176 on PithPith's reviewarXiv
  12. [32] Eric Jang, Alex Irpan, Mohi Khansari, Daniel Kap- pler, Frederik Ebert, Corey Lynch, Sergey Levine, and Chelsea Finn. BC-Z: Zero-Shot Task General- ization with Robotic Imitation Learning. InCon- ference on Robot Learning, volume 164 ofProceed- ings of Machine Learning Researc... on Pith
  13. [33] Joel Jang, Seonghyeon Ye, Zongyu Lin, Jiannan Xiang, Johan Bjorck, Yu Fang, Fengyuan Hu, Spencer Huang, Kaushil Kundalia, Yen-Chen Lin, Loic Magne, Ajay Mandlekar, Avnish Narayan, You Liang Tan, Guanzhi Wang, Jing Wang, Qi Wang, Yinzhen Xu, Xiaohui Zeng, Kaiyuan Zheng, Ruijie ... on PithPith's reviewarXiv
  14. [34] Yunfan Jiang, Agrim Gupta, Zichen Zhang, Guanzhi Wang, Yongqiang Dou, Yanjun Chen, Li Fei-Fei, Anima Anandkumar, Yuke Zhu, and Linxi Fan. VIMA: General Robot Manipula- tion with Multimodal Prompts.arXiv preprint arXiv:2210.03094, 2022. https://arxiv.org/abs/ 2210.03094 on PithPith's reviewarXiv
  15. [35] Alexander Khazatsky, Karl Pertsch, Suraj Nair, Ashwin Balakrishna, Sudeep Dasari, Siddharth Karamcheti, Soroush Nasiriany, Mohan Kumar Srirama, Lawrence Yunliang Chen, Kirsty Ellis, Peter David Fagan, Joey Hejna, Masha Itkina, Marion Lepert, Yecheng Jason Ma, Patrick Tree Mill... on Pith
  16. [36] Moo Jin Kim, Karl Pertsch, Siddharth Karamcheti, Ted Xiao, Ashwin Balakrishna, Suraj Nair, Rafael Rafailov, Ethan Paul Foster, Pannag R. Sanketi, Quan Vuong, Thomas Kollar, Benjamin Burchfiel, Russ Tedrake, Dorsa Sadigh, Sergey Levine, Percy Liang, and Chelsea Finn. OpenVLA: A... on Pith
  17. [37] Vidhi Jain, Maria Attarian, Nikhil J. Joshi, Ayzaan Wahid, Danny Driess, Quan Vuong, Pannag R. Sanketi, Pierre Sermanet, Stefan Welker, Chris- tine Chan, Igor Gilitschenski, Yonatan Bisk, and Debidatta Dwibedi. Vid2Robot: End-to-end Video- conditioned Policy Learning with Cros... on Pith
  18. [38] doi: 10.15607/RSS.2024.XX.052. https: //doi.org/10.15607/RSS.2024.XX.052 on Pithdoi:10.15607/rss.2024.xx.052
  19. [39] Qixiu Li, Yaobo Liang, Zeyu Wang, et al. CogACT: A foundational vision-language-action model for synergizing cognition and action in robotic manip- ulation.arXiv preprint arXiv:2411.19650, 2024 on PithPith's reviewarXiv
  20. [40] Xinghang Li, Peiyan Li, Long Qian, Minghuan Liu, Dong Wang, Jirong Liu, Bingyi Kang, Xiao Ma, Xinlong Wang, Di Guo, Tao Kong, Hanbo Zhang, and Huaping Liu. What matters in building vision- language-action models for generalist robots.arXiv preprint arXiv:2412.14058, 2024. http... on PithPith's reviewarXiv
  21. [41] Xinghang Li, Minghuan Liu, Hanbo Zhang, Cun- jun Yu, Jie Xu, Hongtao Wu, Chilam Cheang, Ya Jing, Weinan Zhang, Huaping Liu, Hang Li, and Tao Kong. Vision-Language Foundation Mod- els as Effective Robot Imitators. InInternational Conference on Learning Representations. OpenRe- ... on Pith
  22. [42] Yi Li, Yuquan Deng, Jesse Zhang, Joel Jang, Mar- ius Memmel, Caelan Reed Garrett, Fabio Ramos, Dieter Fox, Anqi Li, Abhishek Gupta, and Ankit Goyal. HAMSTER: Hierarchical Action Models For Open-World Robot Manipulation. InInternational Conference on Learning Representations. O... on Pith
  23. [43] Zekai Li, Jiayi Yu, Jinlu Zhang, Luyang Niu, Zekun Qi, Jun Zheng, Xuchuan Chen, Dairu Liu, Yintian- run Zhang, He Wang, and Li Yi. Where is physical RSI capability from? an observational scaling law of LLM-as-policy, 2026. https://arxiv.org/abs/ 2610.26848 on PithPith's reviewarXiv
  24. [44] Moo Jin Kim, Chelsea Finn, and Percy Liang. Fine-Tuning Vision-Language-Action Models: Op- timizing Speed and Success.arXiv preprint arXiv:2502.19645, 2025. https://arxiv.org/abs/ 2502.19645 on PithPith's reviewarXiv
  25. [45] Moo Jin Kim, Yihuai Gao, Tsung-Yi Lin, Yen-Chen Lin, Yunhao Ge, Grace Lam, Percy Liang, Shuran Song, Ming-Yu Liu, Chelsea Finn, and Jinwei Gu. Cosmos Policy: Fine-Tuning Video Models for Vi- suomotor Control and Planning.arXiv preprint arXiv:2601.16163, 2026. https://arxiv.org... on PithPith's reviewarXiv
  26. [46] Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matthew Le. Flow match- ing for generative modeling. InInternational Conference on Learning Representations. OpenRe- view.net, 2023.https://openreview.net/pdf?id= PqvMRDCJT9t on Pith
  27. [47] Bo Liu, Yifeng Zhu, Chongkai Gao, Yihao Feng, Qiang Liu, Yuke Zhu, and Peter Stone. LIBERO: Benchmarking knowledge transfer for lifelong robot learning. InAdvances in Neural Information Pro- cessing Systems (Datasets and Benchmarks Track), volume 36, 2023. https://arxiv.org/ab... on Pith
  28. [48] Jiaming Liu, Mengzhen Liu, Zhenyu Wang, Pengju An, Xiaoqi Li, Kaichen Zhou, Senqiao Yang, Renrui Zhang, Yandong Guo, and Shanghang Zhang. Robo- Mamba: Efficient Vision-Language-Action Model for Robotic Reasoning and Manipulation. InAd- vances in Neural Information Processing S... on Pith
  29. [49] Songming Liu, Lingxuan Wu, Bangguo Li, Hengkai Tan, Huayu Chen, Zhengyi Wang, Ke Xu, Hang Su, and Jun Zhu. RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation. InInternational Conference on Learning Representations. OpenRe- view.net, 2025. https://openreview.net/for... on Pith
  30. [50] Corey Lynch, Ayzaan Wahid, Jonathan Tomp- son, Tianli Ding, James Betker, Robert Baruch, Travis Armstrong, and Pete Florence. Interac- 13 tive Language: Talking to Robots in Real Time. arXiv preprint arXiv:2210.06407, 2022. https: //arxiv.org/abs/2210.06407 on PithPith's reviewarXiv
  31. [51] Jacky Liang, Wenlong Huang, Fei Xia, Peng Xu, Karol Hausman, Brian Ichter, Pete Florence, and Andy Zeng. Code as policies: Language model programs for embodied control. InIEEE In- ternational Conference on Robotics and Automa- tion, pages 9493–9500. IEEE, 2023. doi: 10.1109/ I... on Pith
  32. [52] Weifeng Lin, Siyuan Huang, Hao Li, Tingwei Chen, Ruichuan An, Xinyu Wei, Jianbo Liu, and Hong- sheng Li. Rethinking VLM representation for VLA initialization.arXiv preprint arXiv:2605.25802, 2026.https://arxiv.org/abs/2605.25802 on PithPith's reviewarXiv
  33. [53] Suraj Nair, Aravind Rajeswaran, Vikash Kumar, Chelsea Finn, and Abhinav Gupta. R3M: A uni- versal visual representation for robot manipulation. InConference on Robot Learning, volume 205 of Proceedings of Machine Learning Research, pages 892–909. PMLR, 2022.https://proceedings... on Pith
  34. [54] Soroush Nasiriany, Abhiram Maddukuri, Lance Zhang, Adeet Parikh, Aaron Lo, Abhishek Joshi, Ajay Mandlekar, and Yuke Zhu. RoboCasa: Large- scale simulation of everyday tasks for generalist robots. InRobotics: Science and Systems (RSS), on Pith
  35. [55] Dantong Niu, Yuvan Sharma, Giscard Biamby, Jerome Quenum, Yutong Bai, Baifeng Shi, Trevor Darrell, and Roei Herzig. LLARVA: Vision-Action Instruction Tuning Enhances Robot Learning. In Conference on Robot Learning, volume 270 ofPro- ceedings of Machine Learning Research, pages 3333– on Pith
  36. [56] NVIDIA. GR00T N1.5: An improved open foundation model for generalist humanoid robots. https://research.nvidia.com/labs/gear/ gr00t-n1_5/, 2025. Published June 11, 2025; accessed November 12, 2026 on Pith
  37. [57] NVIDIA. Cosmos-Reason2-2B. Model card, on Pith
  38. [58] Arjun Majumdar, Karmesh Yadav, Sergio Arnaud, Yecheng Jason Ma, Claire Chen, Sneha Silwal, Aryan Jain, Vincent-Pierre Berges, Tingfan Wu, Jay Vakil, Pieter Abbeel, Jitendra Malik, Dhruv Batra, Yixin Lin, Oleksandr Maksymets, Aravind Rajeswaran, and Franziska Meier. Where are w... on PithPith's reviewarXiv
  39. [59] Ajay Mandlekar, Soroush Nasiriany, Bowen Wen, Iretiayo Akinola, Yashraj Narang, Linxi Fan, Yuke Zhu, and Dieter Fox. MimicGen: A Data Generation System for Scalable Robot Learning using Human Demonstrations. InConference on Robot Learning, volume 229 ofProceedings of Machine L... on Pith
  40. [60] Open X-Embodiment Collaboration. Open X- embodiment: Robotic learning datasets and RT-X models.arXiv preprint arXiv:2310.08864, 2023. https://arxiv.org/abs/2310.08864 on PithPith's reviewarXiv
  41. [61] Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy V. Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mido Assran, Nicolas Ballas, Wojciech Galuba, Rus- sell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rab... on Pith
  42. [62] https://robocasa.ai/assets/robocasa_ rss24.pdf on Pith
  43. [63] Physical Intelligence. π0.7: A steerable general- ist robotic foundation model with emergent capa- bilities. Technical report, 2026. https://www.pi. website/download/pi07.pdf on Pith
  44. [64] Physical Intelligence, Kevin Black, Noah Brown, et al. π0.5: a vision-language-action model with open-world generalization.arXiv preprint arXiv:2504.16054, 2025 on PithPith's reviewarXiv
  45. [65] Scott E. Reed, Konrad Zolna, Emilio Parisotto, Sergio Gómez Colmenarejo, Alexander Novikov, Gabriel Barth-Maron, Mai Gimenez, Yury Sulsky, Jackie Kay, Jost Tobias Springenberg, Tom Eccles, Jake Bruce, Ali Razavi, Ashley Edwards, Nicolas Heess, Yutian Chen, Raia Hadsell, Oriol ... on Pith
  46. [66] Jinghuan Shang, Karl Schmeckpeper, Brandon B. May, Maria Vittoria Minniti, Tarik Kelestemur, David Watkins, and Laura Herlant. Theia: Distill- ing diverse vision foundation models for robot learn- ing. InConference on Robot Learning, volume 270 ofProceedings of Machine Learnin... on Pith
  47. [67] https://huggingface.co/nvidia/ Cosmos-Reason2-2B on Pith
  48. [68] NVIDIA. Isaac GR00T N1.7: A founda- tion model for generalist robots. Code and model release, 2026.https://github.com/NVIDIA/ Isaac-GR00T. Version N1.7; model card at https://huggingface.co/nvidia/GR00T-N1.7-3B on Pith
  49. [69] NVIDIA. GR00T 2: Scaling robot foun- dation models for generalization, dexter- ity, and adaptation. White paper, 2026. https://research.nvidia.com/labs/gear/ gr00t-2/assets/gr00t2_whitepaper.pdf on Pith
  50. [70] Mustafa Shukor, Dana Aubakirova, Francesco Ca- puano, et al. SmolVLA: A vision-language-action model for affordable and efficient robotics.arXiv preprint arXiv:2506.01844, 2025 on PithPith's reviewarXiv
  51. [71] Ishika Singh, Valts Blukis, Arsalan Mousavian, Ankit Goyal, Danfei Xu, Jonathan Tremblay, Dieter Fox, Jesse Thomason, and Animesh Garg. Prog- Prompt: Generating Situated Robot Task Plans using Large Language Models. InIEEE Inter- national Conference on Robotics and Automation,... on Pith
  52. [72] Karl Pertsch, Kyle Stachowicz, Brian Ichter, Danny Driess, SurajNair, QuanVuong, OierMees, Chelsea Finn, and Sergey Levine. FAST: Efficient Action Tokenization for Vision-Language-Action Models. arXiv preprint arXiv:2501.09747, 2025.https:// arxiv.org/abs/2501.09747 on PithPith's reviewarXiv
  53. [73] Homer Rich Walke, Kevin Black, Tony Z. Zhao, Quan Vuong, Chongyi Zheng, Philippe Hansen- Estruch, Andre Wang He, Vivek Myers, Moo Jin Kim, Max Du, Abraham Lee, Kuan Fang, Chelsea Finn, and Sergey Levine. BridgeData V2: A Dataset for Robot Learning at Scale. InConfer- ence on R... on Pith
  54. [74] Hanwen Wang, Weizhi Zhao, Xiangyu Wang, Siyuan Huang, He Lin, Boyuan Zheng, Rong- tao Xu, Gang Wang, Yao Mu, He Wang, Lue Fan, Hongsheng Li, Zhaoxiang Zhang, and Tie- niu Tan. DexJoCo: A benchmark and toolkit for task-oriented dexterous manipulation on Mu- JoCo.arXiv preprint ... on PithPith's reviewarXiv
  55. [75] Lirui Wang, Xinlei Chen, Jialiang Zhao, and Kaim- ing He. Scaling Proprioceptive-Visual Learning with Heterogeneous Pre-trained Transformers. In Advances in Neural Information Processing Sys- tems, volume 37, 2024. https://arxiv.org/abs/ 2409.20537 on PithPith's reviewarXiv
  56. [76] Yihao Wang, Pengxiang Ding, Lingxiao Li, Can Cui, Zirui Ge, Xinyang Tong, Wenxuan Song, Han Zhao, Wei Zhao, Pengxu Hou, Siteng Huang, Yi- fan Tang, Wenhui Wang, Ru Zhang, Jianyi Liu, and Donglin Wang. VLA-Adapter: An effec- tive paradigm for tiny-scale vision-language-action m... on Pithdoi:10.1609/aaai.v40i22.38931
  57. [77] Yichao Shen, Fangyun Wei, Zhiying Du, Yaobo Liang, Yan Lu, Jiaolong Yang, Nanning Zheng, and Baining Guo. VideoVLA: Video Generators Can Be Generalizable Robot Manipulators. InAdvances in Neural Information Processing Systems, volume 38, 2025.https://arxiv.org/abs/2512.06963 on PitharXiv
  58. [78] Mohit Shridhar, Lucas Manuelli, and Dieter Fox. CLIPort: What and Where Pathways for Robotic Manipulation. InConference on Robot Learning, volume 164 ofProceedings of Ma- chine Learning Research, pages 894–906. PMLR, on Pith
  59. [79] Tete Xiao, Ilija Radosavovic, Trevor Darrell, and Jitendra Malik. Masked Visual Pre-training for Motor Control.arXiv preprint arXiv:2203.06173, 2022.https://arxiv.org/abs/2203.06173 on PithPith's reviewarXiv
  60. [80] Mohit Shridhar, Lucas Manuelli, and Dieter Fox. Perceiver-Actor: A Multi-Task Trans- former for Robotic Manipulation. InConfer- ence on Robot Learning, volume 205 ofProceed- ings of Machine Learning Research, pages 785–799. PMLR, 2022. https://proceedings.mlr.press/ v205/shrid... on Pith
  61. [81] Jinhui Ye, Ning Gao, Senqiao Yang, Jinliang Zheng, Zixuan Wang, Yuxin Chen, Pengguang Chen, Yilun Chen, Shu Liu, and Jiaya Jia. StarVLA-α: Reduc- ing Complexity in Vision-Language-Action Systems. arXiv preprint arXiv:2604.11757, 2026. https: //arxiv.org/abs/2604.11757 on PithPith's reviewarXiv
  62. [82] Seonghyeon Ye, Yunhao Ge, Kaiyuan Zheng, Shenyuan Gao, Sihyun Yu, George Kurian, Suneel Indupuru, You Liang Tan, Chuning Zhu, Jian- nan Xiang, Ayaan Malik, Kyungmin Lee, William 15 Liang, Nadun Ranawaka, Jiasheng Gu, Yinzhen Xu, Guanzhi Wang, Fengyuan Hu, Avnish Narayan, Jo- h... on PithPith's reviewarXiv
  63. [83] Luoyang Sun, Guoyang Xia, Fengfa Li, Lei Ren, Xinyu Cui, Haifeng Zhang, Fangxiang Feng, Kaike Zhang, Kun Zhan, Yan Xie, Jun Wang, and Cheng Deng. What makes an efficient VLA? navigating action-head design, scaling, and latency.arXiv preprint arXiv:2609.13984, 2026. https://arx... on PithPith's reviewarXiv
  64. [84] Haoqi Yuan, Zhixuan Liang, Anzhe Chen, Ye Wang, Haoyang Li, Pei Lin, Yiyang Huang, Zixing Lei, Tong Zhang, Jiazhao Zhang, Jie Zhang, Jingyang Fan, Gengze Zhou, Qihang Peng, Chenxu Lv, Xi- aoyue Chen, An Yang, Fei Huang, Junyang Lin, Dayiheng Liu, Jingren Zhou, Chenfei Wu, and ... on PithPith's reviewarXiv
  65. [85] Wentao Yuan, Jiafei Duan, Valts Blukis, Wilbert Pumacay, Ranjay Krishna, Adithyavairavan Murali, Arsalan Mousavian, and Dieter Fox. RoboPoint: A vision-language model for spatial affordance predic- tion for robotics.arXiv preprint arXiv:2406.10721, 2024.https://arxiv.org/abs/2... on PithPith's reviewarXiv
  66. [86] Michal Zawalski, William Chen, Karl Pertsch, Oier Mees, Chelsea Finn, and Sergey Levine. Robotic control via embodied chain-of-thought reasoning. InConference on Robot Learning, volume 270 of Proceedings of Machine Learning Research, pages 3157–3181. PMLR, 2024. https://procee... on Pith
  67. [87] Yanjie Ze, Gu Zhang, Kangning Zhang, Chenyuan Hu, Muhan Wang, and Huazhe Xu. 3D Diffusion Policy: Generalizable Visuomotor Policy Learning viaSimple3DRepresentations. InRobotics: Science and Systems, 2024. doi: 10.15607/RSS.2024.XX.067. https://doi.org/10.15607/RSS.2024.XX.067 on Pithdoi:10.15607/rss.2024.xx.067
  68. [88] Junjie Wen, Yichen Zhu, Jinming Li, Minjie Zhu, Kun Wu, Zhiyuan Xu, Ning Liu, Ran Cheng, Chaomin Shen, Yaxin Peng, Feifei Feng, and Jian Tang. TinyVLA: Towards Fast, Data-Efficient Vision-Language-Action Models for Robotic Ma- nipulation.arXiv preprint arXiv:2409.12514, 2024. ... on PithPith's reviewarXiv
  69. [89] Hongtao Wu, Ya Jing, Chilam Cheang, Guangzeng Chen, Jiafeng Xu, Xinghang Li, Minghuan Liu, Hang Li, and Tao Kong. Unleashing large-scale video generative pre-training for visual robot ma- nipulation. InInternational Conference on Learn- ing Representations. OpenReview.net, 202... on Pith
  70. [90] Wenyao Zhang, Hongsi Liu, Zekun Qi, Yunnan Wang, Xinqiang Yu, Jiazhao Zhang, Runpei Dong, Jiawei He, He Wang, Zhizheng Zhang, Li Yi, Wen- jun Zeng, and Xin Jin. DreamVLA: A vision- language-action model dreamed with comprehen- sive world knowledge. InAdvances in Neural In- for... on PithPith's reviewarXiv
  71. [91] Jenai Xuning Yang, Ashna Khetan, and Shun Zhang. Astranomy: An empirical study of GPT-6 Astra as a native robot policy. CoRL 2026 Late- Breaking Results,https://research.nvidia.com/ labs/srl/projects/astranomy/, 2026. Published October 14, 2026; accessed November 12, 2026 on Pith
  72. [92] Yuyang Zhang, Wenyao Zhang, Zekun Qi, He Zhang, Haitao Lin, Jingbo Zhang, Yao Mu, Xiaokang Yang, Wenjun Zeng, and Xin Jin. Image- WAM: Do world action models really need video generation, or just image editing? InConfer- ence on Robot Learning (CoRL), 2026. https: //arxiv.org/... on PithPith's reviewarXiv
  73. [93] Qingqing Zhao, Yao Lu, Moo Jin Kim, Zipeng Fu, Zhuoyang Zhang, Yecheng Wu, Zhaoshuo Li, Qianli Ma, Song Han, Chelsea Finn, Ankur Handa, Tsung-Yi Lin, Gordon Wetzstein, Ming-Yu Liu, and Donglai Xiang. CoT-VLA: Visual chain-of-thought reasoning for vision-language-action models.... on Pith
  74. [94] Qize Yu, Lianrui Fan, Boyu Chen, Jiaqi Liang, Xini Ding, Yue Chen, Zetian Song, Yuran Wang, Yi Zou, Kaixuan Wang, Tianxing Chen, Wenxuan Song, Bohan Zhou, Mingleyang Li, Siqiao Huang, Yuqi Ye, CaigaoJiang, WeiWei, RuihaiWu, HangZhang, Yixiao Ge, Shuchang Zhou, Shilong Liu, Xia... on PithPith's reviewarXiv
  75. [95] Haoyu Zhen, Xiaowen Qiu, Peihao Chen, Jincheng Yang, Xin Yan, Yilun Du, Yining Hong, and Chuang Gan. 3D-VLA: A 3D Vision-Language- Action Generative World Model. InInterna- tional Conference on Machine Learning. OpenRe- view.net, 2024. https://openreview.net/forum? id=EZcFK8HupF on Pith
  76. [96] Jinliang Zheng, Jianxiong Li, Zhihao Wang, 16 Dongxiu Liu, Xirui Kang, Yuchun Feng, Yi- nan Zheng, Jiayin Zou, Yilun Chen, Jia Zeng, Ya-Qin Zhang, Jiangmiao Pang, Jingjing Liu, Tai Wang, and Xianyuan Zhan. X-VLA: Soft-Prompted Transformer as Scalable Cross- Embodiment Vision-L... on PithPith's reviewarXiv
  77. [97] Ruijie Zheng, Yongyuan Liang, Shuaiyi Huang, Jianfeng Gao, Hal Daumé III, Andrey Kolobov, Furong Huang, and Jianwei Yang. TraceVLA: Vi- sual Trace Prompting Enhances Spatial-Temporal Awareness for Generalist Robotic Policies. InInter- national Conference on Learning Representa... on Pith
  78. [98] Enshen Zhou, Jingkun An, Cheng Chi, Yi Han, Shanyu Rong, Chi Zhang, Pengwei Wang, Zhongyuan Wang, Tiejun Huang, Lu Sheng, and Shanghang Zhang. RoboRefer: Towards spatial referring with reasoning in vision-language mod- els for robotics. InAdvances in Neural Informa- tion Proce... on PitharXiv
  79. [99] Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, and Lucas Beyer. Sigmoid Loss for Language Image Pre-Training. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 11941–11952. IEEE, 2023. doi: 10. 1109/ICCV51070.2023.01100. https://doi.org/ 10.1... on Pithdoi:10.1109/iccv51070.2023.01100
  80. [100] Jianke Zhang, Xiaoyu Chen, Qiuyue Wang, Ming- sheng Li, Yanjiang Guo, Yucheng Hu, Jiajun Zhang, Shuai Bai, Junyang Lin, and Jianyu Chen. VLM4VLA: Revisiting vision-language-models in vision-language-action models.arXiv preprint arXiv:2601.03309, 2026. https://arxiv.org/abs/ 2601.03309 on PithPith's reviewarXiv

The first 100 of 113 references are listed here. All of them are in this paper's citation record.

Challenges and discussion

Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.

No comments yet.

Sign in to comment