Robotics · arXiv:2609.12641 v1 · submitted 11 Sep 2026 · 37 references
Breaking the Vision-Action Shortcut: Latent Interface Training for Generalizable Robotics Foundation Models
The paper says
“Robot foundation models achieve strong in-distribution performance but often degrade under visual distribution shifts. When learning to generate actions from pretrained visual representations, models may exploit task-irrelevant visual cues that correlate with demonstrated actions within the training distribution.”
more of the abstract
Such vision-action shortcuts can undermine generalization when these correlations change under distribution shifts. Mitigating these shortcuts requires constraining how visual information is used for action generation while preserving task-relevant spatial information. We propose Latent Interface Training (LIT), a framework-agnostic two-stage strategy that first establishes a spatial-goal-conditioned action prior without images, then constrains visual conditioning through a pose-supervised latent interface. Stage 1 trains the action expert to generate action chunks conditioned on language, robot state, and each demonstrated chunk's terminal SE(3) end-effector pose, learning goal-directed action generation independently of visual cues. Stage 2 introduces a latent interface that aggregates visual and semantic representations and serves as the pretrained action expert's only visual conditioning pathway. The interface is supervised to reconstruct the terminal pose previously used to condition Stage 1, encouraging it to retain the goal-relevant spatial information needed for action generation. Across four vision-language-action and world-action architectures (Pi0.5, MolmoAct2, FAST-WAM, and ImageWAM), LIT improves overall LIBERO-Plus success by 3.87-10.70 percentage points while preserving or improving average LIBERO success. Real-world evaluations show 13.30-16.70 percentage-point gains in success aggregated across three tasks under unseen camera configurations, lighting variations, and distractors.
from the abstract
Pith says
Machine review · deepseek-v4-flash · 14 Sep 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, with conditions. LIT proposes a clever two-stage training recipe for reducing vision-action shortcuts, but fixed-seed evaluations and unclear baseline fine-tuning make the headline numbers less solid than they appear. This line is wrong
3 major objections · 6 minor
How often objections from reviews like this one are right: not yet measured
Not checked: figures, code, what the 37 references say.
The paper says, at Section IV-A and Table II
The review did not quote this passage.
Pith says: Objection 1 of 3 · major
Statistical rigor is absent from the central empirical claim. All LIBERO-Plus evaluations use a single fixed evaluation seed with one rollout per instance (Section IV-A). No confidence intervals, standard errors, or multi-seed results are reported. The claimed improvements of 3.87–10.70 percentage points could plausibly fall within the noise floor given the variance across perturbation dimensions.
more of objection 1
This is a load-bearing methodological gap: without measures of variability, the quantitative claim that LIT outperforms baselines is unsupported. The authors should repeat evaluations with at least three seeds and report means with standard deviations or confidence intervals.
The paper says, at Section III-C and Table III
The review did not quote this passage.
Pith says: Objection 2 of 3 · major
The paper asserts that supervising the latent interface to reconstruct the terminal pose (Eq. 8) forces retention of goal-relevant visual information. However, the ablation removing pose supervision (LIT w/o pose supervision) shows only a 3.06 pp drop in overall OOD success, and the variant with pose supervision alone (Baseline w/ pose supervision) improves only 1.83 pp over the baseline.
more of objection 2
These modest effects leave open the possibility that the latent tokens could satisfy the pose reconstruction objective by relying on language and state information alone, while discarding most visual content. The mechanism claim requires evidence that the latent tokens encode visual spatial information beyond what is available from non-visual modalities; e.g., an analysis of whether pose can be predicted from language/state alone, or an information-theoretic measure of visual content retained in the tokens.
The paper says, at Section IV-D and Figure 6
The review did not quote this passage.
Pith says: Objection 3 of 3 · major
Real-robot evaluations use only 10 rollouts per task per OOD condition (25 per task for ID). For a binary success metric, this sample size yields wide binomial confidence intervals (e.g., for 10 rollouts, the 95% CI spans roughly ±30 percentage points for observed rates near 50%).
more of objection 3
The paper reports point estimates without any uncertainty quantification. The authors should report binomial confidence intervals or, at minimum, individual trial outcomes. The claim of 13.30–16.70 percentage-point improvements is not statistically grounded at current sample sizes.
6 minor objections
The paper says, at Abstract and Section I
Pith says: minor objection 1 of 6
The model name appears as 'π0.5' in the abstract but as 'π 0.5' elsewhere. Please use a consistent typographic convention throughout.
The paper says, at Section IV-A
Pith says: minor objection 2 of 6
The phrase 'one rollout per instance with a fixed evaluation seed' is ambiguous. Clarify whether the same seed is used for all rollouts (i.e., no stochasticity in the environment or policy) and whether the seed is held constant across architecture comparisons.
show 4 more
The paper says, at Figure 4
Pith says: minor objection 3 of 6
The caption describes 'Effective action-to-image attention' for LIT, but the attention is composed across two steps (action→latent and latent→visual). The comparability of these composed maps to the baseline direct attention maps is not discussed. Add a note on how the composition is computed.
The paper says, at Section IV-E2 and Figure 7
Pith says: minor objection 4 of 6
The claim 'LIT achieves a lower action loss in 20K visual training steps than the baseline reaches in 30K steps' compares losses at different training steps. Since the total number of gradient updates is matched (30K total), but Stage 2 starts from Stage 1 parameters, the comparison is not apples-to-apples. A fairer comparison would evaluate the baseline at 20K steps as well.
The paper says, at Table II
Pith says: minor objection 5 of 6
The 'Overall' row is described as the unweighted mean over seven perturbations. The weights of each perturbation dimension in the LIBERO-Plus benchmark should be stated explicitly to confirm reproducibility.
The paper says, at Section IV-A
Pith says: minor objection 6 of 6
The sentence 'We do not fine-tune pretrained VLA or WAM policy checkpoints' contains a typo. Additionally, the decision not to use pretrained policy checkpoints means that the reported baseline results are lower than published results for the same architectures. The authors should explicitly state this limitation and discuss how it affects comparability with prior work.
What would settle it
Repeat the entire LIBERO-Plus evaluation (Table II) for each of the four architectures (π0.5, MolmoAct2, FAST-WAM, ImageWAM) using at least 3 independent random seeds (covering action expert initialization, training, and evaluation randomness). For each seed, recompute baseline and LIT success rates for all seven perturbation dimensions and the overall score. Report the mean and standard deviation across seeds, and compute Cohen's d for the baseline–LIT difference. If the mean improvement for any architecture fails to exceed 2× the pooled standard deviation, or if d < 0.5, the claimed gains lack statistical and practical significance.
from the skeptic's reading
How the review went
4 readings by deepseek-v4-flash · 14 Sep 2026
- Reader holds up, with conditions
- Skeptic partly agrees
- Editor wrote the verdict above
- Referee major revision: 3 major objections, 6 minor
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
I'd say this paper is worth a serious look, but don't take the exact numbers at face value. The core idea—first learn an action prior conditioned on terminal SE(3) poses without images, then route vision exclusively through a latent interface supervised to reconstruct that same pose—is genuinely new and cleanly motivated. The cross-architecture experiments (π0.5, MolmoAct2, FAST-WAM, ImageWAM) show consistent gains on LIBERO-Plus, and the ablations do a nice job separating contributions. I also appreciate the real-robot sanity check, even if small.
The soft spots are load-bearing enough that I'd want to see fixes before accepting the central quantitative claim. First: all simulation results come from a single evaluation seed with no confidence intervals. The paper reports gains of 3.87–10.70 percentage points across 28 architecture–perturbation cells. Without multi-seed variability, I can't tell whether those gains exceed the noise floor. Given 10,030 instances the standard error might be small, but a single seed doesn't prove it. Second: the real-robot study uses only 10 rollouts per condition—fine for a feasibility demo, not for claiming 13–17 percentage point improvements. Third: the paper doesn't clearly state whether the baselines also fine-tune the backbone. LIT Stage 2 employs full fine-tuning; if the baselines freeze the backbone, some fraction of the gain could come from simply updating backbone weights. The ablations (e.g., 'direct visual access' still fine-tunes the backbone) partially address this, but it's not ruled out. Fourth: no code or data are released.
These issues are addressable rather than fatal. The method is coherent, the evaluation design is otherwise solid, and the motivation is well-supported. If the authors add multi-seed experiments with variance estimates, clarify the training configurability, and release code, the paper would be a solid conference contribution. As is, I'd treat the gains as promising but unconfirmed.
This paper is for researchers working on generalizable robot foundation models. It deserves peer review—but a diligent one that pushes for statistical rigor.
The strongest objection
The paper's central claim is quantitative: LIT improves LIBERO-Plus overall success by 3.87–10.70 pp across four architectures (Table II). However, all evaluations use a single fixed seed (Section IV-A: 'one rollout per instance with a fixed evaluation seed'). No confidence intervals, standard errors, or multi-seed results are reported. For an evaluation covering 10,030 instances, the standard error of the mean depends on the variance across seeds and perturbations; a fixed seed yields only one draw from the distribution. The real-robot study similarly uses only 10–25 rollouts per condition. If the observed improvements lie within the noise floor of the baseline, the central empirical claim is unsupported. Ablations (Table III) also lack variance information, making it impossible to assess whether component removals produce statistically distinguishable drops. Given that the paper's novelty hinges on these numerical gains, this methodological gap is the most load-bearing concern: without statistical rigor, we cannot confidently accept that LIT outperforms baselines reliably.
The reader identified statistical weaknesses (no variance, fixed seed, small real-robot sample) as secondary issues while focusing the weakest_assumption on the pose-supervision mechanism. I consider the statistical concern more load-bearing because the paper's headline quantitative claims (the 3.87–10.70 pp improvements) are the primary evidence for LIT's effectiveness. If those numbers are not statistically robust, the entire empirical foundation is weakened, regardless of the mechanism's plausibility. The paper can address this concern with additional runs, making it addressable and thus keeping the verdict at CONDITIONAL. I partially agree with the reader because they noted the issue but did not treat it as the weakest assumption.
The referee's report
If the results are robust, the paper introduces a conceptually clean mechanism—pose-supervised latent interface as the sole visual conditioning pathway—that could reduce reliance on spurious visual correlations and improve generalization of robot foundation models. The approach is architecture-agnostic and demonstrated across four recent designs, which suggests potential for broad applicability. However, the empirical contribution is currently undermined by a lack of statistical rigor, which tempers the significance.
Pith's reading of the paper (a machine's summary, not the authors')
This paper argues that a major cause of poor generalization under visual distribution shifts in robot foundation models is the vision-action shortcut: models learn to rely on task-irrelevant visual cues that correlate with actions in the training data. To break this shortcut, the authors propose Latent Interface Training (LIT), a two-stage strategy that first learns a goal-directed action prior without any images, then introduces visual conditioning exclusively through a latent interface supervised to reconstruct the same spatial goal. The central claim is that this forces the interface to retain only goal-relevant spatial information while discarding shortcut-prone visual patterns. If true, LIT would offer a model-agnostic way to make existing VLA and WAM architectures substantially more robust to changes in viewpoint, lighting, distractors, and other visual perturbations without sacrificing in-distribution performance.
What carries the argument. The key machinery is a two-stage training pipeline. Stage 1 trains the action expert from scratch using a flow-matching objective conditioned on language, robot state, and the terminal SE(3) end-effector pose of each action chunk, but crucially without any visual input—this learns a spatial-goal-conditioned action prior. Stage 2 freezes the action expert and introduces learnable latent tokens that cross-attend to both visual and semantic backbone representations at each layer. These latent tokens become the action expert's only visual conditioning pathway, and they are supervised via an MLP decoder to reconstruct the same terminal pose used in Stage 1. This shared pose target enforces that the interface retains goal-relevant spatial information while filtering out shortcut-prone visual correlations.
Core claim. The paper discovers that training a robot foundation model's action expert first without images (conditioned only on language, robot state, and each action chunk's terminal SE(3) end-effector pose), then introducing visual conditioning exclusively through a latent interface that is supervised to reconstruct the same terminal pose, yields consistent generalization gains across four distinct VLA and WAM architectures (π0.5, MolmoAct2, FAST-WAM, ImageWAM). On the LIBERO-Plus zero-shot generalization benchmark, LIT improves overall success by 3.87–10.70 percentage points across architectures, with gains especially large under camera viewpoint changes (up to +27.43 points) and sensor noise. These improvements transfer to real-robot tasks with 13.30–16.70 percentage-point gains under lighting, camera, and distractor shifts. Ablations show that removing any component—the image-free action prior, the pose supervision, or the exclusive visual pathway—significantly reduces LIT's gains, confirming that the combination is necessary.
The premise it rests on. The terminal SE(3) pose reconstruction objective in Stage 2 is assumed to be sufficient to force the latent interface to retain exactly the visual spatial information needed for action generation, and that this pose cannot be predicted from language and robot state alone; if the pose can be predicted trivially without visual content, the interface could discard all visual information and still minimize the pose loss, breaking the claimed shortcut mitigation.
If the paper is right (Pith's inferences, not the authors')
If LIT works as claimed, training robot foundation models on any existing demonstration dataset can be augmented with this two-stage procedure to improve robustness without collecting more diverse data.
The method may be applicable beyond manipulation tasks to any visuomotor policy where a spatial goal (such as a target end-effector pose) can be extracted from the demonstration trajectory.
The success of LIT suggests that explicitly decoupling spatial-goal learning from visual appearance learning is a principled route to generalization, potentially reducing the need for large-scale data augmentation or domain randomization.
LIT's consistent improvements across very different coupling mechanisms (shared self-attention vs. layer-wise cross-attention) indicate that the bottleneck principle—routing all visual conditioning through a spatially supervised interface—may be a general design pattern for robust robot learning.
A direct extension not explored in the paper: LIT could be combined with action-chunking schemes that predict multiple future poses, not just the terminal pose, to provide richer spatial targets for the latent interface and possibly yield even better generalization.
The paper's ablation showing that adding pose supervision alone to the baseline (without the interface) gives only a 1.83-point gain suggests that the supervision must be paired with a bottleneck that forces visual information through the interface—this implies the interface is doing more than just providing pose information; it is structuring how visual features are used.
An implicit assumption of the method is that the terminal SE(3) pose is a sufficient goal signal for action generation. If a task requires visual information that cannot be reduced to this pose (e.g., recognizing object fragility or texture to adjust grasp force), LIT might discard necessary information and underperform. Testing on such tasks would reveal a boundary condition.
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
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.
The paper
Abstract
Robot foundation models achieve strong in-distribution performance but often degrade under visual distribution shifts. When learning to generate actions from pretrained visual representations, models may exploit task-irrelevant visual cues that correlate with demonstrated actions within the training distribution. Such vision-action shortcuts can undermine generalization when these correlations change under distribution shifts. Mitigating these shortcuts requires constraining how visual information is used for action generation while preserving task-relevant spatial information. We propose Latent Interface Training (LIT), a framework-agnostic two-stage strategy that first establishes a spatial-goal-conditioned action prior without images, then constrains visual conditioning through a pose-supervised latent interface. Stage 1 trains the action expert to generate action chunks conditioned on language, robot state, and each demonstrated chunk's terminal SE(3) end-effector pose, learning goal-directed action generation independently of visual cues. Stage 2 introduces a latent interface that aggregates visual and semantic representations and serves as the pretrained action expert's only visual conditioning pathway. The interface is supervised to reconstruct the terminal pose previously used to condition Stage 1, encouraging it to retain the goal-relevant spatial information needed for action generation. Across four vision-language-action and world-action architectures (Pi0.5, MolmoAct2, FAST-WAM, and ImageWAM), LIT improves overall LIBERO-Plus success by 3.87-10.70 percentage points while preserving or improving average LIBERO success. Real-world evaluations show 13.30-16.70 percentage-point gains in success aggregated across three tasks under unseen camera configurations, lighting variations, and distractors.
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 · 2026-09-23
The authors' main claim, as the classifier picked it out 0.80
“Across four vision-language-action and world-action architectures (Pi0.5, MolmoAct2, FAST-WAM, and ImageWAM), LIT improves overall LIBERO-Plus success by 3.87-10.70 percentage points while preserving or improving average LIBERO success.”
How the main claim is stated: asserted 1.00
What the abstract itself states, as probabilities: claims first 0.05 · states limitation 0.13 · human participants 0.06 · controlled comparison 0.66 · randomized assignment 0.02 · states quantitative result 0.99 · compares with prior results 0.76 · 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.
| Sentence | Role | The authors' own work | States the main conclusion | Strength |
|---|---|---|---|---|
| Robot foundation models achieve strong in-distribution performance but often degrade under visual distribution shifts. | background 0.97 | 0.09 | 0.03 | asserted 0.96 |
| When learning to generate actions from pretrained visual representations, models may exploit task-irrelevant visual cues that correlate with demonstrated actions within the training distribution. | background 0.95 | 0.10 | 0.03 | hedged possibility 1.00 |
| Such vision-action shortcuts can undermine generalization when these correlations change under distribution shifts. | background 0.93 | 0.06 | 0.05 | asserted 0.54 |
| Mitigating these shortcuts requires constraining how visual information is used for action generation while preserving task-relevant spatial information. | background 0.73 | 0.06 | 0.08 | asserted 0.97 |
| We propose Latent Interface Training (LIT), a framework-agnostic two-stage strategy that first establishes a spatial-goal-conditioned action prior without images, then constrains visual conditioning through a pose-supervised latent interface. | contribution 0.80 | 0.87 | 0.20 | asserted 1.00 |
| Stage 1 trains the action expert to generate action chunks conditioned on language, robot state, and each demonstrated chunk's terminal SE(3) end-effector pose, learning goal-directed action generation independently of visual cues. | method 0.99 | 0.89 | 0.04 | asserted 1.00 |
| Stage 2 introduces a latent interface that aggregates visual and semantic representations and serves as the pretrained action expert's only visual conditioning pathway. | method 0.98 | 0.83 | 0.05 | asserted 1.00 |
| The interface is supervised to reconstruct the terminal pose previously used to condition Stage 1, encouraging it to retain the goal-relevant spatial information needed for action generation. | method 1.00 | 0.84 | 0.06 | asserted 0.92 |
| Across four vision-language-action and world-action architectures (Pi0.5, MolmoAct2, FAST-WAM, and ImageWAM), LIT improves overall LIBERO-Plus success by 3.87-10.70 percentage points while preserving or improving average LIBERO success. | result 1.00 | 0.96 | 0.48 | asserted 1.00 |
| Real-world evaluations show 13.30-16.70 percentage-point gains in success aggregated across three tasks under unseen camera configurations, lighting variations, and distractors. | result 1.00 | 0.95 | 0.30 | asserted 1.00 |
What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 0.97
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, 2026-09-23: 1 request, 7,706 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 (7)
Showing 6 of the paper's 7 figures. Figures are the authors', reproduced from the paper.
References (37)
- [1] K. Black, N. Brown, J. Darpinian, K. Dhabalia, D. Driesset al., “π 0.5: A vision-language-action model with open-world generalization,” in Proceedings of the 9th Conference on Robot Learning (CoRL), ser. Proceedings of Machine Learning Research, vol. 305. PMLR, 2025, pp. 17–40 on Pith
- [2] H. Fang, J. Duan, D. Clay, S. Wang, S. Liuet al., “MolmoAct2: Action reasoning models for real-world deployment,”arXiv preprint arXiv:2605.02881, 2026 on PithPith's reviewarXiv
- [3] T. Yuan, Z. Dong, Y . Liu, and H. Zhao, “Fast-W AM: Do world action models need test-time future imagination?”arXiv preprint arXiv:2603.16666, 2026 on PithPith's reviewarXiv
- [4] Y . Zhang, W. Zhang, Z. Qi, H. Zhang, H. Linet al., “ImageW AM: Do world action models really need video generation, or just image editing?”arXiv preprint arXiv:2606.19531, 2026 on PithPith's reviewarXiv
- [5] C. Chi, S. Feng, Y . Du, Z. Xu, E. Cousineauet al., “Diffusion policy: Visuomotor policy learning via action diffusion,” inProceedings of Robotics: Science and Systems (RSS), Daegu, Republic of Korea, Jul. 2023 on Pith
- [6] K. Black, N. Brown, D. Driess, A. Esmail, M. Equiet al., “π 0: A vision-language-action flow model for general robot control,” in Proceedings of Robotics: Science and Systems (RSS), Los Angeles, CA, USA, Jun. 2025 on Pith
- [7] S. Fei, S. Wang, J. Shi, Z. Dai, J. Caiet al., “LIBERO-Plus: A progressive robustness benchmark for visual-language-action models,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Jun. 2026, pp. 38 574–38 583 on Pith
- [8] R. Geirhos, J.-H. Jacobsen, C. Michaelis, R. Zemel, W. Brendel et al., “Shortcut learning in deep neural networks,”Nature Machine Intelligence, vol. 2, no. 11, pp. 665–673, Nov. 2020 on Pith
- [9] D. Qu, H. Song, Q. Chen, Y . Yao, X. Yeet al., “SpatialVLA: Exploring spatial representations for visual-language-action models,” inProceedings of Robotics: Science and Systems (RSS), Los Angeles, CA, USA, Jun. 2025 on Pith
- [10] J. Lee, J. Duan, H. Fang, Y . Deng, S. Liuet al., “MolmoAct: Action reasoning models that can reason in space,”arXiv preprint arXiv:2508.07917, 2025 on PithPith's reviewarXiv
- [11] R. Zheng, Y . Liang, S. Huang, J. Gao, H. Daum´e IIIet al., “TraceVLA: Visual trace prompting enhances spatial-temporal awareness for gen- eralist robotic policies,” inInternational Conference on Learning Representations (ICLR), 2025 on Pith
- [12] T. Lin, Y . Du, Y . Mao, Z. Ye, Y . Zhonget al., “LA4VLA: Learning to act without seeing via language-action pretraining,”arXiv preprint arXiv:2606.27295, 2026 on PithPith's reviewarXiv
- [13] Q. Wang, M. Li, J. Guan, J. Ye, S. Xieet al., “Qwen-VLA: Unifying vision-language-action modeling across tasks, environments, and robot embodiments,”arXiv preprint arXiv:2605.30280, 2026 on PithPith's reviewarXiv
- [14] B. Liu, Y . Zhu, C. Gao, Y . Feng, Q. Liuet al., “LIBERO: Bench- marking knowledge transfer for lifelong robot learning,” inAdvances in Neural Information Processing Systems (NeurIPS) Datasets and Benchmarks Track, vol. 36, 2023 on Pith
- [15] A. O’Neillet al., “Open X-Embodiment: Robotic learning datasets and RT-X models,” in2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 6892–6903 on Pith
- [16] D. Ghosh, H. R. Walke, K. Pertsch, K. Black, O. Meeset al., “Octo: An open-source generalist robot policy,” inProceedings of Robotics: Science and Systems (RSS), Delft, Netherlands, Jul. 2024 on Pith
- [17] B. Zitkovich, T. Yu, S. Xu, P. Xu, T. Xiaoet al., “RT-2: Vision- language-action models transfer web knowledge to robotic control,” inProceedings of the 7th Conference on Robot Learning (CoRL), ser. Proceedings of Machine Learning Research, vol. 229. PMLR, 2023, pp. 2165–2183 on Pith
- [18] M. J. Kim, K. Pertsch, S. Karamcheti, T. Xiao, A. Balakrishna et al., “OpenVLA: An open-source vision-language-action model,” in Proceedings of the 8th Conference on Robot Learning (CoRL), ser. Proceedings of Machine Learning Research, vol. 270. PMLR, 2025, pp. 2679–2713 on Pith
- [19] Q. Li, Y . Liang, Z. Wang, L. Luo, X. Chenet al., “CogACT: A foundational vision-language-action model for synergizing cognition and action in robotic manipulation,”arXiv preprint arXiv:2411.19650, 2024 on PithPith's reviewarXiv
- [20] S. Ye, Y . Ge, K. Zheng, S. Gao, S. Yuet al., “World action models are zero-shot policies,”arXiv preprint arXiv:2602.15922, 2026 on PithPith's reviewarXiv
show 17 more
- [21] D. Driess, J. T. Springenberg, B. Ichter, L. Yu, A. Li-Bellet al., “Knowledge insulating vision-language-action models: Train fast, run fast, generalize better,”arXiv preprint arXiv:2505.23705, 2025 on PithPith's reviewarXiv
- [22] P. de Haan, D. Jayaraman, and S. Levine, “Causal confusion in imitation learning,” inAdvances in Neural Information Processing Systems (NeurIPS), vol. 32. Curran Associates, Inc., 2019, pp. 11 693–11 704 on Pith
- [23] X. Li, K. Hsu, J. Gu, O. Mees, K. Pertschet al., “Evaluating real- world robot manipulation policies in simulation,” inProceedings of the 8th Conference on Robot Learning (CoRL), ser. Proceedings of Machine Learning Research, vol. 270. PMLR, 2025, pp. 3705–3728 on Pith
- [24] S. Zhang, Z. Xu, P. Liu, X. Yu, Y . Liet al., “VLABench: A large-scale benchmark for language-conditioned robotics manipulation with long-horizon reasoning tasks,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), Oct. 2025, pp. 11 142–11 152 on Pith
- [25] S. Nair, A. Rajeswaran, V . Kumar, C. Finn, and A. Gupta, “R3M: A universal visual representation for robot manipulation,” inProceedings of the 6th Conference on Robot Learning (CoRL), ser. Proceedings of Machine Learning Research, vol. 205. PMLR, 2023, pp. 892–909 on Pith
- [26] T. Xiao, I. Radosavovic, T. Darrell, and J. Malik, “Masked visual pre- training for motor control,”arXiv preprint arXiv:2203.06173, 2022 on PithPith's reviewarXiv
- [27] M. Zawalski, W. Chen, K. Pertsch, O. Mees, C. Finn, and S. Levine, “Robotic control via embodied chain-of-thought reasoning,” inPro- ceedings of the 8th Conference on Robot Learning (CoRL), ser. Proceedings of Machine Learning Research, vol. 270. PMLR, 2025, pp. 3157–3181 on Pith
- [28] Q. Zhao, Y . Lu, M. J. Kim, Z. Fu, Z. Zhanget al., “CoT-VLA: Visual chain-of-thought reasoning for vision-language-action models,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Jun. 2025, pp. 1702–1713 on Pith
- [29] Y . Li, Y . Deng, J. Zhang, J. Jang, M. Memmelet al., “HAMSTER: Hierarchical action models for open-world robot manipulation,” in International Conference on Learning Representations (ICLR), 2025 on Pith
- [30] D. Hwang, B. Lee, D. Kim, H. Jang, H. Jinet al., “3D HAMSTER: Bridging planning and control in hierarchical vision language action models through 3D trajectory guidance,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2026 on Pith
- [31] X. Li, V . Belagali, J. Shang, and M. S. Ryoo, “Crossway diffu- sion: Improving diffusion-based visuomotor policy via self-supervised learning,” in2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 16 841–16 849 on Pith
- [32] J. Park, Y . Seo, C. Liu, L. Zhao, T. Qinet al., “Object-aware regularization for addressing causal confusion in imitation learning,” inAdvances in Neural Information Processing Systems (NeurIPS), vol. 34. Curran Associates, Inc., 2021, pp. 3029–3042 on Pith
- [33] A. Eftekhar, K.-H. Zeng, J. Duan, A. Farhadi, A. Kembhavi, and R. Krishna, “Selective visual representations improve convergence and generalization for embodied AI,” inInternational Conference on Learning Representations (ICLR), 2024 on Pith
- [34] D. Jing, T. Zhang, J. Liu, J. Zhao, Z. Sun, L. E. Li, Z. Lu, and M. Ding, “Learning action priors for cross-embodiment robot manipulation,” arXiv preprint arXiv:2606.26095, 2026 on PithPith's reviewarXiv
- [35] K. Xu, Z. Zhu, A. Chen, R. Xiong, and Y . Wang, “APT: Action expert pretraining improves instruction generalization of vision-language- action policies,”arXiv preprint arXiv:2606.12366, 2026 on PithPith's reviewarXiv
- [36] Y . Lipman, R. T. Q. Chen, H. Ben-Hamu, M. Nickel, and M. Le, “Flow matching for generative modeling,” inInternational Conference on Learning Representations (ICLR), 2023 on Pith
- [37] Y . Wang, P. Ding, L. Li, C. Cui, Z. Geet al., “VLA-Adapter: An effective paradigm for tiny-scale vision-language-action model,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 40, no. 22, 2026 on Pith
Cited by 3 Pith papers
Reviewed papers on Pith that cite this one. The full citation record
- Gnomon4D: Learning Where and When to Focus for Geometric Motion-Grounded World-Action Modeling cs.CV · Nov 2026
- Acting Through Registers cs.RO · Oct 2026
- TacitVLA: Reasoning Vision-Language-Action Models Without Reasoning Latency cs.AI · Oct 2026
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.
Come back by email
A short list of the week's newly reviewed papers. Nothing is sent until you confirm from your inbox, and every email has a link that stops it. No account.
No comments yet.
Sign in to comment