Pith.

Artificial Intelligence · arXiv:2610.17540 v1 · submitted 13 Oct 2026 · 66 references

TacitVLA: Reasoning Vision-Language-Action Models Without Reasoning Latency

Jack Lu, Yurong You, Boyi Li, Yulong Cao, Thomas Tian, Chaowei Xiao, Xiangyu Chen, Wenjie Luo

show 3 more authors

Boris Ivanovic, Marco Pavone, Yan Wang

The paper says

“Textual reasoning before acting has improved vision-language-action (VLA) models, but at the cost of latency: every action conditioned on the reasoning text must wait for autoregressive decoding.”

more of the abstract

We show that actions do not need to wait for or condition on the generated reasoning to benefit from it, because the reasoning supervision already improves the backbone representation. We introduce TacitVLA, a reasoning VLA whose action expert never attends to the reasoning it generates for the current call. On long-horizon tasks where the reasoning text also serves as episodic memory, it can generate each entry during action execution rather than before acting. We evaluate it in a controlled study on four benchmarks spanning manipulation and autonomous driving, memory and non-memory tasks: RoboMME, RMBench, LIBERO, and PhysicalAI-AV. On all four, TacitVLA matches the performance of its explicit-reasoning counterpart at up to 3.4x lower latency, and matches the latency of the same architecture trained without reasoning at up to +17.1% success. In absolute terms, it surpasses the best previously published results on the RoboMME Counting suite (82.7% vs. 66.8%) and RMBench (77.5% vs. 56.5%), and reaches 72.9% on LIBERO-Plus and a 52.5% scene score on the public split of the PhysicalAI-AV closed-loop driving challenge. Project page: https://tacitvla.github.io.

from the abstract

Pith says

Machine review · deepseek-v4-flash · 14 Oct 2026 · what was read was not recorded · not checked by a person

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

Holds up. Clean controlled study confirms that reasoning supervision alone is enough—test-time generation is unnecessary for performance. Practical advance with honest limitations. This line is wrong

3 minor

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

Are you an author? What you can do.

Not checked: figures, code, what the 66 references say.

3 minor objections

The paper says, at Appendix C, Table 15

Pith says: minor objection 1 of 3

The execution window for RMBench is defined as the motion planner's mean time per plan. Because this is a simulation artifact and can vary between calls, stating that 'no deferred generation exceeded its window in any call' would be strengthened by reporting the maximum observed generation time and the minimum observed execution window, rather than only the means.

The paper says, at Abstract

Pith says: minor objection 2 of 3

The phrase 'up to 3.4× lower latency' could be misinterpreted; it refers to non-memory tasks. The paper later clarifies this distinction in Section 6.2. Consider adding a brief qualifier in the abstract for precision.

show 1 more

The paper says, at Section 6.1, Table 3 (RoboMME all tasks)

Pith says: minor objection 3 of 3

On the counting suite, TacitVLA (82.7±3.8) and Explicit Reasoning (81.5±1.8) have overlapping error bars; the claim that they 'match' is appropriate. On the full 16-task set, TacitVLA (48.1±1.0) and Explicit Reasoning (45.8±1.5) also overlap. The text could note that the standard deviations are from three seeds, which is standard, but the confidence intervals are wide enough that small differences are not excluded.

What would settle it

Train TacitVLA on a non-memory benchmark (e.g., LIBERO) where the reasoning annotations are corrupted (e.g., shuffled per-frame, or replaced with generic scene descriptions that are consistent within each episode but not specific to the task). If the performance gap over No Reasoning persists, it would strengthen the claim that any consistent extra text supervision (not necessarily reasoning) improves the backbone; if it narrows, it would confirm that reasoning-specific content is required. Either outcome informs future work without contradicting the paper's core finding.

from the skeptic's reading

How the review went

4 readings by deepseek-v4-flash · 14 Oct 2026

  1. Reader holds up confidence high
  2. Skeptic agrees 57 s · 30,247 tokens read, 5,414 written
  3. Editor wrote the verdict above 15 s · 30,927 tokens read, 1,248 written
  4. Referee accept: 3 minor 18 s · 30,474 tokens read, 1,973 written

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

Pith did not check this review's objections against the paper.

The review in full

The review in brief, machine-written

This paper does what you wish every reasoning-VLA paper would do: a single controlled change (attention mask), everything else held fixed, across four benchmarks. The result is straightforward and convincing: an action expert that never conditions on the current reasoning text matches one that does, at much lower latency. The engineering is solid—they measure latency carefully, report per-stage breakdowns, and handle memory tasks with deferred generation that fit inside execution windows. The comparison to prior work is fair; they correctly note that earlier studies (ECoT-Lite, ZR-0) varied multiple factors at once, so their own ablation is clearer.

The soft spots are acknowledged, which I appreciate. The biggest is that the result depends on the quality of the reasoning data used for supervision—if the supervisory signal is weak or noisy, the advantage could shrink. The authors state this. On memory tasks, the deferred generation adds no latency only when it fits in the execution window; they quantify it at ≤23% of the window in their settings, but this is a parameter that could bind in other domains. The absolute comparisons against published state-of-the-art are not perfectly matched (different backbones, training data, parameter counts), but the paper’s core claim rests on its internal control, which is fair.

There are no methodological red flags. The latency numbers are concrete, the attention-mask change is minimal and well-motivated, and the results are consistent across four very different benchmarks (manipulation, driving, memory, non-memory). This is a paper that a serious editor should send to peer review. It will likely be a useful reference for practitioners building real-time reasoning VLAs, even though it does not change the theoretical understanding of reasoning.

For your reading group: yes, it’s worth discussing as a model of clean empirical work. I would cite it in my own writing on VLA latency. The authors are clear thinkers.

The strongest objection

No significant objection identified. The paper presents a clean controlled study with consistent results across four diverse benchmarks. The central claim—that an action expert never conditioning on the current reasoning text matches the performance of one that does—is well supported by the experimental design, which varies only the action expert's attention mask while holding all other training factors fixed. The acknowledged limitation concerning reasoning data quality is appropriately stated and does not undermine the primary finding for the datasets tested. The latency measurements are concrete and the deferred-generation strategy on memory tasks is validated empirically.

The reader identified the quality and coverage of reasoning data as the weakest assumption, which the paper also states as a limitation. My review confirms that this is the most plausible vulnerability, but the paper's empirical results are robust and the experiments are well-controlled. No additional methodological flaws or logical gaps were found. The proposed concrete test probes this vulnerability further, but even if it revealed sensitivity, it would qualify rather than refute the central claim. Hence the reader's ACCEPT verdict remains appropriate.

The referee's report

If the result holds, it has strong practical significance: it decouples the training-time benefit of reasoning supervision from the deployment-time cost of reasoning generation, enabling lower-latency VLA policies without sacrificing performance. The controlled experimental design—varying only the attention mask while holding training data, prompts, and optimizer fixed—makes the attribution clean. The findings also contribute evidence that the primary benefit of reasoning in these architectures comes from representation learning during training, not from test-time conditioning on generated text. The latency measurements are concrete and reproduce across benchmarks.

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

The paper establishes that reasoning supervision during training improves the backbone representation so much that the action expert does not need to attend to the generated reasoning text at deployment. This removes the latency bottleneck of autoregressive reasoning generation while matching the success of models that require it. On memory tasks where reasoning text serves as episodic memory, generation can run concurrently with action execution instead of before it. If true, any reasoning VLA can be trained with reasoning targets but deployed without the reasoning latency, achieving both high performance and fast response.

What carries the argument. The key machinery is the attention mask that isolates the action expert from the reasoning token positions. By training on the same sequence but masking the action expert's attention to the reasoning text (a single cell in the attention mask), TacitVLA decouples action prediction from text generation. This allows the action expert to condition only on the context (observations, instruction, ledger) while the backbone is still trained to generate reasoning text via cross-entropy loss.

Core claim. The paper discovers that an action expert that never conditions on the current reasoning text matches the performance of one that does, across four benchmarks spanning manipulation and autonomous driving, memory and non-memory tasks. This shows that the benefit of reasoning reaches the action primarily through the supervision signal improving backbone representations, not through the generated text as an input feature. On non-memory tasks, reasoning text generation can be omitted entirely; on memory tasks, it can be deferred to run during action execution.

The premise it rests on. The reasoning data used for supervision is of sufficient quality and coverage to improve backbone representations; the paper acknowledges this as a limitation.

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

Any existing reasoning VLA with a separate action expert can be converted to TacitVLA by masking reasoning tokens, immediately reducing latency without retraining from scratch.

On memory tasks, the ledger entry can be generated during the execution window of the preceding action chunk, so reasoning adds no latency as long as generation finishes within that window.

The reasoning supervision improves performance even when no text is generated at deployment, suggesting that reasoning data acts as a regularizer or representation enhancer rather than a necessary input.

On non-memory tasks, the reasoning text can be generated post-hoc for interpretability without affecting control latency.

This result suggests that chain-of-thought reasoning in embodied AI may primarily serve as a training signal that shapes internal representations, analogous to how rationale distillation improves language models without requiring the rationale at test time.

The finding opens the possibility of training VLAs on synthetic reasoning data generated by larger models, then deploying without generating that reasoning – potentially scaling up reasoning supervision without scaling deployment compute.

For safety-critical applications like autonomous driving, the ability to generate reasoning text during action execution (asynchronously) provides a human-interpretable log without compromising real-time control.

Check this review

Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.

How often reviews like this one are right

Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.

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

Textual reasoning before acting has improved vision-language-action (VLA) models, but at the cost of latency: every action conditioned on the reasoning text must wait for autoregressive decoding. We show that actions do not need to wait for or condition on the generated reasoning to benefit from it, because the reasoning supervision already improves the backbone representation. We introduce TacitVLA, a reasoning VLA whose action expert never attends to the reasoning it generates for the current call. On long-horizon tasks where the reasoning text also serves as episodic memory, it can generate each entry during action execution rather than before acting. We evaluate it in a controlled study on four benchmarks spanning manipulation and autonomous driving, memory and non-memory tasks: RoboMME, RMBench, LIBERO, and PhysicalAI-AV. On all four, TacitVLA matches the performance of its explicit-reasoning counterpart at up to 3.4x lower latency, and matches the latency of the same architecture trained without reasoning at up to +17.1% success. In absolute terms, it surpasses the best previously published results on the RoboMME Counting suite (82.7% vs. 66.8%) and RMBench (77.5% vs. 56.5%), and reaches 72.9% on LIBERO-Plus and a 52.5% scene score on the public split of the PhysicalAI-AV closed-loop driving challenge. Project page: https://tacitvla.github.io.

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 · 2026-10-14

The authors' main claim, as the classifier picked it out 0.88

“We show that actions do not need to wait for or condition on the generated reasoning to benefit from it, because the reasoning supervision already improves the backbone representation.”

How the main claim is stated: asserted 1.00

What the abstract itself states, as probabilities: claims first 0.07 · states limitation 0.17 · human participants 0.05 · controlled comparison 0.95 · randomized assignment 0.02 · states quantitative result 0.99 · compares with prior results 0.98 · mentions code or data availability 0.34

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
Textual reasoning before acting has improved vision-language-action (VLA) models, but at the cost of latency: every action conditioned on the reasoning text must wait for autoregressive decoding. background 0.98 0.08 0.05 asserted 1.00
We show that actions do not need to wait for or condition on the generated reasoning to benefit from it, because the reasoning supervision already improves the backbone representation. interpretation 0.40 0.87 0.80 asserted 1.00
We introduce TacitVLA, a reasoning VLA whose action expert never attends to the reasoning it generates for the current call. method 0.57 0.84 0.19 asserted 1.00
On long-horizon tasks where the reasoning text also serves as episodic memory, it can generate each entry during action execution rather than before acting. method 0.87 0.64 0.08 hedged possibility 0.79
We evaluate it in a controlled study on four benchmarks spanning manipulation and autonomous driving, memory and non-memory tasks: RoboMME, RMBench, LIBERO, and PhysicalAI-AV. method 0.70 0.93 0.03 asserted 0.96
On all four, TacitVLA matches the performance of its explicit-reasoning counterpart at up to 3.4x lower latency, and matches the latency of the same architecture trained without reasoning at up to +17.1% success. result 1.00 0.96 0.50 asserted 1.00
In absolute terms, it surpasses the best previously published results on the RoboMME Counting suite (82.7% vs. 66.8%) and RMBench (77.5% vs. 56.5%), and reaches 72.9% on LIBERO-Plus and a 52.5% scene score on the public split of the PhysicalAI-AV closed-loop driving challenge. result 1.00 0.91 0.22 asserted 1.00
Project page: https://tacitvla.github.io. contribution 0.56 0.10 0.01 unclear 0.98

What kind of paper the classifier reports this is: method or tool 1.00 · 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, 2026-10-14: 1 request, 6,499 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

Figures (8)
Figure 1 of the paper
Figure 1. TacitVLA retains the benefit of reasoning supervision without conditioning its actions on the generated text. Left: how the policies act in one call. Given the context 𝑐𝑡, on memory tasks including the ledger 𝑀𝑡 of past subtasks, the backbone generates the reasoning 𝑟𝑡 and conditions the action expert that outputs the action chunk 𝑎𝑡. In Explicit Reasoning, the action expert attends to 𝑟𝑡, so every action chunk waits for it. TacitVLA trains on the same sequence with 𝑟𝑡 masked from the action expert, bypassing the need to generate 𝑟𝑡 on non-memory tasks. On memory tasks, 𝑟𝑡 still…
Figure 2 of the paper
Figure 2. Training and deployment of Explicit Reasoning and TacitVLA. The two policies share the training sequence and differ in one attention-mask cell, which decouples TacitVLA’s actions from text generation at deployment. (a) The backbone predicts the reasoning 𝑟𝑡 and the tokenized actions 𝑎˜𝑡 (an auxiliary target) under cross-entropy, and the action expert denoises the action chunk 𝑎𝑡 under flow matching. (b) Rows attend to columns. The backbone is causal. The action expert attends to 𝑐𝑡 and its own action chunk, to 𝑟𝑡 only in Explicit Reasoning, and never to 𝑎˜𝑡. The crossed cell is that…
Figure 3 of the paper
Figure 3. TacitVLA matches the latency of No Reasoning on every benchmark, and Explicit Reasoning’s latency is 1.4× to 3.4× higher. Each bar is one policy’s latency per call: 𝑇action for No Reasoning and TacitVLA, and 𝑇action + 𝑇reason for Explicit Reasoning, whose actions wait for text generation. The ratio above each group is Explicit Reasoning’s latency over TacitVLA’s. On the memory tasks, TacitVLA generates the ledger entry during execution, taking at most 23% of the execution window, so the next call never waits for generation. All values are means over model calls on one H100, with the…
Figure 4 of the paper
Figure 4. RoboMME BinFill. The front camera at the calls that generate a new ledger entry and at the episode’s last call. Counting-suite policies of
Figure 5 of the paper
Figure 5. RMBench Press Button. The head camera at the calls that generate a new ledger entry and at the episode’s last call. A call can add two entries when a press completes within the action chunk. Policies of
Figure 6 of the paper
Figure 6. LIBERO. The agent camera at six calls, with the opening of the reasoning generated at each. Policies of
Figure 7 of the paper
Figure 7. LIBERO-Plus, Camera Viewpoints. The agent camera on two tasks under perturbed viewpoints. Policies of
Figure 8 of the paper
Figure 8. PhysicalAI-AV closed loop. The front camera at five calls of two public-split scenes, with the chain of causation generated at each call. Policies of

Figures are the authors', reproduced from the paper.

References (66)
  1. [1] Shuai Bai, Yuxuan Cai, Ruizhe Chen, Keqin Chen, Xionghui Chen, Zesen Cheng, Lianghao Deng, Wei Ding, Chang Gao, Chunjiang Ge, Wenbin Ge, Zhifang Guo, Qidong Huang, Jie Huang, Fei Huang, Binyuan Hui, Shutong Jiang, Zhaohai Li, Mingsheng Li, Mei Li, Kaixin Li, Zicheng Lin, Junya... on PithPith's reviewarXiv
  2. [2] Shuanghao Bai, Jing Lyu, Wanqi Zhou, Zhe Li, Dakai Wang, Lei Xing, Xiaoguang Zhao, Pengwei Wang, Zhongyuan Wang, Cheng Chi, Badong Chen, and Shanghang Zhang. Latent Reasoning VLA: Latent Thinking and Prediction for Vision-Language-Action Models. InInternational Conference on M... on PithPith's reviewarXiv
  3. [3] Florent Bartoccioni, Elias Ramzi, Victor Besnier, Shashanka Venkataramanan, Tuan-Hung Vu, Yihong Xu, LoickChambon, SpyrosGidaris, SerkanOdabas, DavidHurych, RenaudMarlet, AlexandreBoulch, Mickael Chen, Eloi Zablocki, Andrei Bursuc, Eduardo Valle, and Matthieu Cord. VaViM and V... on PithPith's reviewarXiv
  4. [4] Kevin Black, Noah Brown, James Darpinian, Karan Dhabalia, Danny Driess, Adnan Esmail, Michael Robert Equi, Chelsea Finn, Niccolo Fusai, Manuel Y. Galliker, Dibya Ghosh, Lachy Groom, Karol Hausman, Brian Ichter, Szymon Jakubczak, Tim Jones, Liyiming Ke, Devin LeBlanc, Sergey Le... on PithPith's reviewarXiv
  5. [5] Kevin Black, Noah Brown, Danny Driess, Adnan Esmail, Michael Robert Equi, Chelsea Finn, Niccolo Fusai, Lachy Groom, Karol Hausman, Brian Ichter, Szymon Jakubczak, Tim Jones, Liyiming Ke, Sergey Levine, Adrian Li-Bell, Mohith Mothukuri, Suraj Nair, Karl Pertsch, Lucy Xiaoyang S... on PithPith's reviewarXiv
  6. [6] Kevin Black, Manuel Galliker, and Sergey Levine. Real-Time Execution of Action Chunking Flow Policies. InAdvances in Neural Information Processing Systems (NeurIPS), 2025c. URLhttps://arxiv.org/abs/ 2506.07339. 3 on PithPith's reviewarXiv
  7. [7] Hao Chen, Jiaming Liu, Chenyang Gu, Zhuoyang Liu, Renrui Zhang, Xiaoqi Li, Xiao He, Yandong Guo, Chi-Wing Fu, Shanghang Zhang, and Pheng-Ann Heng. Fast-in-Slow: A Dual-System VLA Model Unifying Fast Manipulation within Slow Reasoning. InAdvances in Neural Information Processin... on PithPith's reviewarXiv
  8. [8] Tianxing Chen, Yuran Wang, Mingleyang Li, Yan Qin, Hao Shi, Zixuan Li, Yifan Hu, Yingsheng Zhang, Kaixuan Wang, Yue Chen, Hongcheng Wang, Junjie Wang, Tianhang Yang, Renjing Xu, Ruihai Wu, Yao Mu, Yaodong Yang, Hao Dong, and Ping Luo. RMBench: Memory-Dependent Robotic Manipula... on PithPith's reviewarXiv
  9. [9] William Chen, Suneel Belkhale, Suvir Mirchandani, Karl Pertsch, Danny Driess, Oier Mees, and Sergey Levine. Training Strategies for Efficient Embodied Reasoning. InConference on Robot Learning ( CoRL), 2025b. URLhttps://arxiv.org/abs/2505.08243. 3 on PithPith's reviewarXiv
  10. [10] Cheng Chi, Siyuan Feng, Yilun Du, Zhenjia Xu, Eric Cousineau, Benjamin CM Burchfiel, and Shuran Song. Diffusion Policy: Visuomotor Policy Learning via Action Diffusion. InRobotics: Science and Systems (RSS), on Pith
  11. [11] Yinpei Dai, Hongze Fu, Jayjun Lee, Yuejiang Liu, Haoran Zhang, Jianing Yang, Chelsea Finn, Nima Fazeli, and Joyce Chai. RoboMME: Benchmarking and Understanding Memory for Robotic Generalist Policies. In International Conference on Machine Learning (ICML), 2026. URL https://arx... on PithPith's reviewarXiv
  12. [12] Dwip Dalal, Shivansh Patel, Chahit Jain, Jeonghwan Kim, Utkarsh Mishra, Alex Baratian, Hyeonjeong Ha, Heng Ji, Svetlana Lazebnik, and Unnat Jain. Generalizable VLA Finetuning via Representation Anchoring and Language-Action Alignment, 2026. URLhttps://arxiv.org/abs/2607.13429. 8, 19 on PithPith's reviewarXiv
  13. [13] Ronghao Dang, Jiayan Guo, Bohan Hou, Sicong Leng, Kehan Li, Xin Li, Jiangpin Liu, Yunxuan Mao, Zhikai Wang, Yuqian Yuan, Minghao Zhu, Xiao Lin, Yang Bai, Qian Jiang, Yaxi Zhao, Minghua Zeng, Junlong Gao, Yuming Jiang, Jun Cen, Siteng Huang, Liuyi Wang, Wenqiao Zhang, Chengju L... on PitharXiv
  14. [14] Yuntian Deng, Kiran Prasad, Roland Fernandez, Paul Smolensky, Vishrav Chaudhary, and Stuart Shieber. Implicit Chain of Thought Reasoning via Knowledge Distillation, 2023. URLhttps://arxiv.org/abs/ 2311.01460. 3 on PithPith's reviewarXiv
  15. [15] Zhekai Duan, Yuan Zhang, Shikai Geng, Gaowen Liu, Joschka Boedecker, and Chris Xiaoxuan Lu. Fast ECoT: Efficient Embodied Chain-of-Thought via Thoughts Reuse. InIEEE International Conference on Robotics and Automation (ICRA), 2026. URLhttps://arxiv.org/abs/2506.07639. 3 11 Tac... on PitharXiv
  16. [16] Haoquan Fang, Markus Grotz, Wilbert Pumacay, Yi Ru Wang, Dieter Fox, Ranjay Krishna, and Jiafei Duan. SAM2Act: Integrating Visual Foundation Model with A Memory Architecture for Robotic Manipulation. In International Conference on Machine Learning (ICML), 2025. URL https://arx... on PithPith's reviewarXiv
  17. [17] Haoquan Fang, Jiafei Duan, Donovan Clay, Sam Wang, Shuo Liu, Weikai Huang, Xiang Fan, Wei-Chuan Tsai, Shirui Chen, Yi Ru Wang, Shanli Xing, Jaemin Cho, Jae Sung Park, Ainaz Eftekhar, Peter Sushko, Karen Farley, Angad Wadhwa, Cole Harrison, Winson Han, Ying-Chun Lee, Eli Vander... on PithPith's reviewarXiv
  18. [18] Senyu Fei, Siyin Wang, Junhao Shi, Zihao Dai, Jikun Cai, Pengfang Qian, Li Ji, Xinzhe He, Shiduo Zhang, Zhaoye Fei, Jinlan Fu, Jingjing Gong, and Xipeng Qiu. LIBERO-Plus: A Progressive Robustness Benchmark for Visual-Language-Action Models. InProceedings of the IEEE/CVF Confer... on PithPith's reviewarXiv
  19. [19] Figure AI. Helix: A Vision-Language-Action Model for Generalist Humanoid Control.https://www. figure.ai/news/helix, 2025. Technical report, February 2025. 2 on Pith
  20. [20] Yuxia Fu, Zhizhen Zhang, Yuqi Zhang, Zijian Wang, Zi Huang, and Yadan Luo. MergeVLA: Cross-Skill Model Merging Toward a Generalist Vision-Language-Action Agent. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition ( CVPR), 2026. URL https://arxiv... on PitharXiv
show 46 more
  1. [21] Chongkai Gao, Zixuan Liu, Zhenghao Chi, Junshan Huang, Xin Fei, Yiwen Hou, Yuxuan Zhang, Yudi Lin, Zhirui Fang, and Lin Shao. VLA-OS: Structuring and Dissecting Planning Representations and Paradigms in Vision-Language-Action Models. InAdvances in Neural Information Processing... on PithPith's reviewarXiv
  2. [22] Hongbo Gao, Zeyu Ni, Xin Wen, Siyu Xu, and Ruifeng Li. AGM: Achievement-Grounded Memory for Closed-Loop Agents with Frozen VLA Policies, 2026. URLhttps://arxiv.org/abs/2608.29537. 7 on PithPith's reviewarXiv
  3. [23] Shibo Hao, Sainbayar Sukhbaatar, DiJia Su, Xian Li, Zhiting Hu, Jason E Weston, and Yuandong Tian. Training Large Language Models to Reason in a Continuous Latent Space. InConference on Language Modeling ( COLM), 2025. URLhttps://arxiv.org/abs/2412.06769. 3 on PithPith's reviewarXiv
  4. [24] Zeyuan He, Bowen Yang, Zhirui Fang, Keru Zhou, Lei Jiang, Jingjing Qian, Fan Mo, Junchi Yan, Philip Torr, Xiu Li, Li Jiang, and Jialin Yu. ELAN4D: Embodiment-Centric 4D Supervision for Vision-Language-Action Models via Plug-and-Play Adaptation, 2026. URLhttps://arxiv.org/abs/2... on PithPith's reviewarXiv
  5. [25] Cheng-Yu Hsieh, Chun-Liang Li, Chih-kuan Yeh, Hootan Nakhost, Yasuhisa Fujii, Alex Ratner, Ranjay Krishna, Chen-Yu Lee, and Tomas Pfister. Distilling Step-by-Step! Outperforming Larger Language Models with Less Training Data and Smaller Model Sizes. InFindings of the Associati... on PithPith's reviewarXiv
  6. [26] Chi-Pin Huang, Yueh-Hua Wu, Min-Hung Chen, Frank Wang, and Fu-En Yang. ThinkAct: Vision-Language- Action Reasoning via Reinforced Visual Latent Planning. InAdvances in Neural Information Processing Systems (NeurIPS), 2025. URLhttps://arxiv.org/abs/2507.16815. 2 on PithPith's reviewarXiv
  7. [27] Chi-Pin Huang, Yunze Man, Zhiding Yu, Min-Hung Chen, Jan Kautz, Yu-Chiang Frank Wang, and Fu-En Yang. Fast-ThinkAct: Efficient Vision-Language-Action Reasoning via Verbalizable Latent Planning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognitio... on PitharXiv
  8. [28] Yuhua Jiang, Junjie Lu, Xinyao Qin, Xiaoyu Chen, Kaixin Wang, Feifei Gao, and Li Zhao. VLA-GSE: Boosting Parameter-Efficient Fine-Tuning in VLA with Generalized and Specialized Experts, 2026. URL https://arxiv.org/abs/2605.06175. 19 on PithPith's reviewarXiv
  9. [29] Donggyun Kim, Jack Lu, Chanwoo Kim, Mengye Ren, and Seunghoon Hong. HyperThink: Text-to- Parameter Hypernetworks for Efficient Reasoning. InConference on Language Modeling ( COLM), 2026. URLhttps://arxiv.org/abs/2610.03039. 3 on PithPith's reviewarXiv
  10. [30] Jason Lee, Jiafei Duan, Haoquan Fang, Yuquan Deng, Shuo Liu, Boyang Li, Bohan Fang, Jieyu Zhang, Yi Ru Wang, Sangho Lee, Winson Han, Wilbert Pumacay, Angelica Wu, Rose Hendrix, Karen Farley, Eli VanderBilt, Ali Farhadi, Dieter Fox, and Ranjay Krishna. MolmoAct: Action Reasonin... on PithPith's reviewarXiv
  11. [31] Haoyang Li, Guanlin Li, Youhe Feng, Chen Zhao, Zhuoran Wang, Yang Li, Qizhe Wei, Shifeng Bao, Haitao Shen, Yihan Zhao, Tong Yang, and Jing Zhang. Training Vision-Language-Action Models with Dense Embodied Chain-of-Thought Supervision, 2026a. URLhttps://arxiv.org/abs/2606.30552. 2, 3 on PithPith's reviewarXiv
  12. [32] Wei Li, Peijin Jia, Yuan Ma, Xuefeng Jiang, Titong Jiang, Sheng Sun, Yujian Li, Xin Wen, Han Hong, Zhikang Liu, Bailin Li, and Kun Zhan. FoMoVLA: Bridging Visual Foresight and Motion Guidance for Vision-Language-Action Models, 2026b. URLhttps://arxiv.org/abs/2607.14739. 19 on PithPith's reviewarXiv
  13. [33] Weixin Liang, Lili Yu, Liang Luo, Srini Iyer, Ning Dong, Chunting Zhou, Gargi Ghosh, Mike Lewis, Wen-tau Yih, Luke Zettlemoyer, and Xi Victoria Lin. Mixture-of-Transformers: A Sparse and Scalable Architecture for Multi-Modal Foundation Models.Transactions on Machine Learning R... on PithPith's reviewarXiv
  14. [34] Fanqi Lin, Ruiqian Nai, Yingdong Hu, Jiacheng You, Junming Zhao, and Yang Gao. OneTwoVLA: A Unified Vision-Language-Action Model with Adaptive Reasoning. InInternational Conference on Learning Representations (ICLR), 2026a. URLhttps://arxiv.org/abs/2505.11917. 2, 3 on PitharXiv
  15. [35] Jianman Lin, Shailesh Shailesh, Zhongyi Luo, and Jiafei Duan. Breaking the Vision-Action Shortcut: Latent Interface Training for Generalizable Robotics Foundation Models, 2026b. URLhttps://arxiv. org/abs/2609.12641. 8, 19 on PithPith's reviewarXiv
  16. [36] Tao Lin, Yuxin Du, Jiting Liu, Nuobei Zhu, Yunhe Li, Yuqian Fu, Yinxinyu Chen, Hongyi Cai, Zewei Ye, Bing Cheng, Kai Ye, Yiran Mao, Yilei Zhong, MingKang Dong, Junchi Yan, Gen Li, and Bo Zhao. Evo-Depth: A Lightweight Depth-Enhanced Vision-Language-Action Model, 2026c. URLhttp... on PithPith's reviewarXiv
  17. [37] Zefu Lin, Rongxu Cui, Junjia Xu, Xiaojuan Jin, Wenling Li, Lue Fan, and Zhaoxiang Zhang. World Pilot: Steering Vision-Language-Action Models with World-Action Priors, 2026d. URLhttps://arxiv.org/ abs/2606.12403. 8, 19 on PithPith's reviewarXiv
  18. [38] Bo Liu, Yifeng Zhu, Chongkai Gao, Yihao Feng, Qiang Liu, Yuke Zhu, and Peter Stone. LIBERO: Bench- marking Knowledge Transfer for Lifelong Robot Learning. InAdvances in Neural Information Processing Systems (NeurIPS), Datasets and Benchmarks Track, 2023. URL https://arxiv.org/... on PithPith's reviewarXiv
  19. [39] Jiaming Liu, Mengzhen Liu, Zhenyu Wang, Pengju An, Xiaoqi Li, Kaichen Zhou, Senqiao Yang, Renrui Zhang, Yandong Guo, and Shanghang Zhang. RoboMamba: Efficient Vision-Language-Action Model for Robotic Reasoning and Manipulation. InAdvances in Neural Information Processing Syste... on Pith
  20. [40] NVIDIA. PhysicalAI Autonomous Vehicles Dataset. Hugging Face dataset, 2025. URL https:// huggingface.co/datasets/nvidia/PhysicalAI-Autonomous-Vehicles. 6 13 TacitVLA: Reasoning Vision-Language-Action Models Without Reasoning Latency on Pith
  21. [41] NVIDIA. AlpaSim E2E Closed Loop Challenge 2026. https://huggingface.co/spaces/nvidia/ AlpasimE2EClosedLoopChallenge2026, 2026. Challenge guide, protocol, and the curated NuRec splits in thee2e_challengebranch ofhttps://github.com/NVlabs/alpasim. 6, 8, 9, 18 on Pith
  22. [42] NVIDIA, Yulong Cao, Riccardo de Lutio, Sanja Fidler, Guillermo Garcia Cobo, Zan Gojcic, Maximilian Igl, Boris Ivanovic, Peter Karkus, Janick Martinez Esturo, Marco Pavone, Aaron Smith, Ellie Tanimura, Michal Tyszkiewicz, Michael Watson, Qi Wu, and Le Zhang. AlpaSim: A Modular,... on Pith
  23. [43] NVIDIA, Yan Wang, Wenjie Luo, Junjie Bai, Yulong Cao, Tong Che, Ke Chen, Yuxiao Chen, Jenna Diamond, Yifan Ding, Wenhao Ding, Liang Feng, Greg Heinrich, Jack Huang, Peter Karkus, Boyi Li, Pinyi Li, Tsung-Yi Lin, Dongran Liu, Ming-Yu Liu, Langechuan Liu, Zhijian Liu, Jason Lu, ... on PithPith's reviewarXiv
  24. [44] Karl Pertsch, Kyle Stachowicz, Brian Ichter, Danny Driess, Suraj Nair, Quan Vuong, Oier Mees, Chelsea Finn, and Sergey Levine. FAST: Efficient Action Tokenization for Vision-Language-Action Models. In Robotics: Science and Systems (RSS), 2025. URLhttps://arxiv.org/abs/2501.09747. 4 on PithPith's reviewarXiv
  25. [45] Boyang Shen, Kaixiang Yang, Hao Wang, Qiuyu Yu, Qiang Xie, Qiang Li, and Zhiwei Wang. LoopVLA: Learning Sufficiency in Recurrent Refinement for Vision-Language-Action Models, 2026. URLhttps: //arxiv.org/abs/2605.09948. 8 on PithPith's reviewarXiv
  26. [46] Hao Shi, Bin Xie, Yingfei Liu, Lin Sun, Fengrong Liu, Tiancai Wang, Erjin Zhou, Haoqiang Fan, Xiangyu Zhang, and Gao Huang. MemoryVLA: Perceptual-Cognitive Memory in Vision-Language-Action Models for Robotic Manipulation. InInternational Conference on Learning Representations ... on PithPith's reviewarXiv
  27. [47] Ajay Sridhar, Jennifer Pan, Satvik Sharma, and Chelsea Finn. MemER: Scaling Up Memory for Robot Control via Experience Retrieval, 2025. URLhttps://arxiv.org/abs/2510.20328. 7 on PitharXiv
  28. [48] Nan Sun, Yuan Zhang, Yongkun Yang, Wentao Zhao, Peiyan Li, Jun Guo, Wenxuan Song, Pengxiang Ding, Runze Suo, Yifei Su, Xin Xiao, Xinghang Li, and Huaping Liu. Revisiting Embodied Chain-of-Thought for Generalizable Robot Manipulation, 2026a. URLhttps://arxiv.org/abs/2606.03784. 2, 3 on PithPith's reviewarXiv
  29. [49] Xiaoquan Sun, Ruijian Zhang, Chen Cao, Yihan Sun, Jiahui Chen, Zetian Xu, Bo Chen, Haijier Chen, Zhen Yang, Jiarun Zhu, Yijun Hong, JingZhe Xu, Jingrui Pang, Mingqi Yuan, and Jiayu Chen. HiMem- WAM: Hierarchical Memory-Gated World Action Models for Robotic Manipulation, 2026b.... on PithPith's reviewarXiv
  30. [50] Shuhan Tan, Kashyap Chitta, Yuxiao Chen, Ran Tian, Yurong You, Yan Wang, Wenjie Luo, Yulong Cao, Philipp Krähenbühl, Marco Pavone, and Boris Ivanovic. Latent Chain-of-Thought World Modeling for End-to-End Autonomous Driving. InProceedings of the IEEE/CVF Conference on Computer... on PithPith's reviewarXiv
  31. [51] Yihao Wang, Pengxiang Ding, Lingxiao Li, Can Cui, Zirui Ge, Xinyang Tong, Wenxuan Song, Han Zhao, Wei Zhao, Pengxu Hou, Siteng Huang, Yifan Tang, Wenhui Wang, Ru Zhang, Jianyi Liu, and Donglin Wang. VLA-Adapter: An Effective Paradigm for Tiny-Scale Vision-Language-Action Model... on PitharXiv
  32. [52] Siyu Xu, Yunke Wang, Chenghao Xia, Dihao Zhu, Tao Huang, and Chang Xu. VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching. InAdvances in Neural Information Processing Systems (NeurIPS), 2025. URLhttps://arxiv.org/abs/2502.02175. 3 on PitharXiv
  33. [53] Ganlin Yang, Zhangzheng Tu, Yuqiang Yang, Sitong Mao, Junyi Dong, Tianxing Chen, Jiaqi Peng, Jing Xiong, Jiafei Cao, Jifeng Dai, Wengang Zhou, Yao Mu, and Tai Wang. EventVLA: Event-Driven Visual Evidence Memory for Long-Horizon Vision-Language-Action Policies, 2026a. URLhttps:... on PithPith's reviewarXiv
  34. [54] Sizhe Yang, Juncheng Mu, Tianming Wei, Chenhao Lu, Xiaofan Li, Linning Xu, Zhengrong Xue, Zhecheng Yuan, Dahua Lin, Jiangmiao Pang, and Huazhe Xu. MemoryWAM: Efficient World Action Modeling with Persistent Memory, 2026b. URLhttps://arxiv.org/abs/2606.20562. 3, 19 on PithPith's reviewarXiv
  35. [55] Yi Yang, Zhihong Liu, Siqi Kou, Yiyang Chen, Yanzhe Hu, Jianbo Zhou, Boyuan Zhao, Zhijie Wei, Xiao Xia, Xueqi Li, Pengfei Liu, and Zhijie Deng. World-Language-Action Model for Unified World Modeling, Language Reasoning, and Action Synthesis, 2026c. URLhttps://arxiv.org/abs/260... on PithPith's reviewarXiv
  36. [56] Cheng Yin, Yankai Lin, Wang Xu, Sikyuen Tam, Xiangrui Zeng, Zhiyuan Liu, and Zhouping Yin. Deep- ThinkVLA: Enhancing Reasoning Capability of Vision-Language-Action Models. InConference on Language Modeling ( COLM), 2026. URLhttps://arxiv.org/abs/2511.15669. 2, 3, 5, 6, 16, 19 on PithPith's reviewarXiv
  37. [57] Tianyuan Yuan, Zibin Dong, Yicheng Liu, and Hang Zhao. Fast-WAM: Do World Action Models Need Test-time Future Imagination?, 2026. URLhttps://arxiv.org/abs/2603.16666. 3, 10 on PithPith's reviewarXiv
  38. [58] Yang Yue, Yulin Wang, Bingyi Kang, Yizeng Han, Shenzhi Wang, Shiji Song, Jiashi Feng, and Gao Huang. DeeR-VLA: Dynamic Inference of Multimodal Large Language Models for Efficient Robot Execution. In Advances in Neural Information Processing Systems (NeurIPS), 2024. URL https:/... on Pith
  39. [59] Michał Zawalski, William Chen, Karl Pertsch, Oier Mees, Chelsea Finn, and Sergey Levine. Robotic Control via Embodied Chain-of-Thought Reasoning. InConference on Robot Learning ( CoRL), 2024. URL https://arxiv.org/abs/2407.08693. 2, 3, 4 on PithPith's reviewarXiv
  40. [60] Jianke Zhang, Yanjiang Guo, Xiaoyu Chen, Yen-Jen Wang, Yucheng Hu, Chengming Shi, and Jianyu Chen. HiRT: Enhancing Robotic Control with Hierarchical Robot Transformers. InConference on Robot Learning ( CoRL), 2024. URLhttps://arxiv.org/abs/2410.05273. 2, 3 on PithPith's reviewarXiv
  41. [61] Wenyao Zhang, Hongsi Liu, Zekun Qi, Yunnan Wang, Xinqiang Yu, Jiazhao Zhang, Runpei Dong, Jiawei He, He Wang, Zhizheng Zhang, Li Yi, Wenjun Zeng, and Xin Jin. DreamVLA: A Vision-Language-Action Model Dreamed with Comprehensive World Knowledge. InAdvances in Neural Information ... on PithPith's reviewarXiv
  42. [62] Tony Z. Zhao, Vikash Kumar, Sergey Levine, and Chelsea Finn. Learning Fine-Grained Biman- ual Manipulation with Low-Cost Hardware. InRobotics: Science and Systems (RSS), 2023. URL https://arxiv.org/abs/2304.13705. 4, 7 on PithPith's reviewarXiv
  43. [63] Jinliang Zheng, Jianxiong Li, Zhihao Wang, Dongxiu Liu, Xirui Kang, Yuchun Feng, Yinan Zheng, Jiayin Zou, Yilun Chen, Jia Zeng, Tai Wang, Ya-Qin Zhang, Jingjing Liu, and Xianyuan Zhan. X-VLA: Soft- Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Model.... on PithPith's reviewarXiv
  44. [64] Shoujing Zhu, Zhenyang Liu, Fungmiu Wang, Jiafeng Wang, Bo Yue, Guiliang Liu, Simo Wu, Xiangyang Xue, and Taiping Zeng. WeaveLA: Event Driven Cross-Subtask Latent Memory Weaving for Repetitive Robot Manipulation, 2026. URLhttps://arxiv.org/abs/2606.17463v1. 7 15 TacitVLA: Reas... on PithPith's reviewarXiv
  45. [2023] URLhttps://arxiv.org/abs/2303.04137. 4, 7 on PithPith's reviewarXiv
  46. [2024] URLhttps://arxiv.org/abs/2406.04339. 3 on PithPith's reviewarXiv
Cited by 2 Pith papers

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