Pith. sign in

REVIEW 3 major objections 5 minor 72 references

Merging and Disentangling Views in Visual Reinforcement Learning for Robotic Manipulation

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read MAD trains multi-view policies that keep working when a camera goes missing.

desk verdict Solid empirical multi-view RL paper; the merging-plus-feature-augmentation combo is new and the evaluation is broad, but the headline margins rest on an alpha tuned on the same tasks, so treat them as conditional. read the letter →

arxiv 2505.04619 v2 pith:AMG4UURQ submitted 2025-05-07 cs.LG cs.CVcs.RO

Abdulaziz Almuzairee , Rohan Patil , Dwait Bhatt , Henrik I. Christensen This is my paper · ORCID
classification cs.LGcs.CVcs.RO
keywords visualreinforcementlearningmulti-viewrobotfeaturedisentanglementdataaugmentationroboticmanipulationsampleefficiencycamerarobustnesssensorfailure
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The central claim is that merging multiple camera views and disentangling them are complementary goals that a single training scheme can achieve. The proposed MAD algorithm encodes each camera view with a shared network, sums the view features into one merged representation for the actor and critic, and also feeds each single-view feature into those networks as a regularizing augmentation. The result, if the paper is right, is a policy that learns faster than methods that only merge or only disentangle views, and that can be deployed with any one camera. This matters because real robot deployments are burdened by multi-camera setups and are sensitive to camera failure. The paper reports higher success rates than four baselines across 20 simulated manipulation tasks, including on singular third-person views.

What carries the argument

The mechanism is feature-level data augmentation built on feature summation. Each view $o_t^i$ is encoded by a shared CNN into $V_t^i$, and summation produces the merged representation $M_t = \sum_i V_t^i$, which keeps the feature dimension identical for one or many views and preserves a signal of how many views contributed. The SADA stabilization recipe is then reused: in both actor and critic updates, targets are predicted only from the unaugmented merged stream $M_t$, while online predictions also see each single-view feature $V_t^i$ as an augmentation, and the two objectives are combined with hyperparameter $\alpha$. This lets the downstream networks learn from both the merged state and every individual view without the unstable multi-state training that the paper's naive-both ablation exhibits.

What would settle it

Run the MAD versus MAD-Naive-Both comparison from Figure 5 on all 20 tasks with a full hyperparameter sweep of $\alpha$ and the same seed budget; if a tuned naive-both baseline reaches MAD's success on both merged and singular views, then the SADA-style target separation is not what carries the claimed robustness, and the central mechanism would be unsupported.

Watch

Extended reading notes

Core claim

MAD's central claim is that merging and disentangling are complementary: training on the summed multi-view feature $M_t = \sum_{i=1}^n V_t^i$ while regularizing the actor and critic with each single-view feature $V_t^i$ as a feature-level augmentation yields both higher sample efficiency and robustness to camera loss. The loss follows the SADA recipe, with learning targets computed only from the unaugmented merged stream while online Q-values and policy actions are also computed from each single-view stream, balanced by an $\alpha$ weighting set to $0.8$. On 15 Meta-World and 5 ManiSkill3 tasks, the paper reports higher average success rates than MVD, VIB, MV-MWM, and single-camera DrQ across all-camera and singular-view evaluations, with the largest margins on the third-person views.

Load-bearing premise

The method assumes that the stabilization recipe designed for image-level augmentation also works when the augmented inputs are single-view feature vectors and the unaugmented input is their sum; if that transfer fails, the training degenerates into the unstable naive-both variant that the paper's own ablation reports.

Editorial extensions

If this is right

  • A single MAD training run yields a policy deployable with any one camera, so the camera array can be simplified at deployment time without retraining or added parameters.
  • MAD does not depend on any particular view: in the occlusion experiment it keeps solving ManiSkill3 tasks when two cameras are pointed at uninformative scenes and only one view is useful.
  • The same merge-and-augment structure transfers across modalities: an RGBD-trained MAD policy can be evaluated with RGB-only or depth-only inputs, and its depth-only success rate exceeds a depth-only DrQ baseline.
  • The gains persist with more cameras: on five views in ManiSkill3, MAD's average success rate is above all baselines, and mixing different input resolutions costs little when images are resized to a common size.

Reading between the lines

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

  • The paper leaves implicit that the recipe may generalize beyond cameras: any set of complementary features (modalities, proprioceptive channels, or subgoal encodings) could be summed and then re-fed individually as augmentations, turning MAD into a generic sensor-dropout training scheme.
  • Because feature summation preserves the magnitude of the summed features, the critic may implicitly learn to count the number of active views; testing whether value estimates shift with the number of summed features could reveal the mechanism behind the graceful degradation.
  • The authors note the absence of real-robot validation; a natural next test is sim-to-real transfer with one physical camera disabled, which would show whether the robustness survives calibration shifts and image noise.
  • The fixed $\alpha=0.8$ suggests the merged stream should dominate but not exclude single views; a task-dependent or adaptive $\alpha$ could improve robustness further on tasks where one view is much more informative than others.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes MAD (Merge And Disentangle), a modification of the DrQ visual RL algorithm for multi-camera robotic manipulation. Each camera view is encoded by a shared CNN and the resulting features are summed into a merged representation; the actor and critic are trained on the merged representation as an unaugmented stream and on each single-view feature as a feature-level augmentation, following the SADA recipe with an additional weighting hyperparameter alpha. The method is evaluated on 15 Meta-World and 5 ManiSkill3 tasks against MVD, VIB, MV-MWM, and Single Camera DrQ, with the central claim that MAD increases sample efficiency while remaining robust to reductions in input camera views. The paper also reports component ablations, alpha ablations, comparisons of merging methods, occlusion experiments, modality experiments, and scalability experiments with five views and mixed input resolutions.

Significance. If the reported results hold, MAD is a simple and lightweight contribution to multi-view visual RL: it needs no auxiliary losses, no additional learnable parameters, no ordered views, and no extra forward passes, while simultaneously improving sample efficiency and robustness to missing views. The empirical breadth is a genuine strength: 20 tasks, 5 seeds, 95% confidence intervals, component ablations, merge-method comparisons, occlusion and modality experiments, and an extended five-view experiment. The main result is also externally benchmarked against strong published baselines rather than only against self-derived variants. The central risks are the undisclosed alpha tuning protocol and the fact that the SADA stabilization recipe is imported from prior work without an ablation that isolates it from the alpha weighting; both are addressable within the manuscript's scope. The absence of real-robot validation is explicitly acknowledged in Section 6 and is not, by itself, a blocker for the stated simulation-based claims.

major comments (3)
  1. [Section 4.3, Eq. (1)-(2), Table 1, Figure 5] The choice alpha=0.8 is load-bearing, but the tuning protocol is not reported. Section 4.3 states only 'after tuning alpha', Table 1 fixes alpha=0.8, and Figure 5 (bottom) shows that alpha=0.5 and alpha=1.0 both average around 26% success on the 5 Meta-World hard tasks while alpha=0.8 reaches 56%. Since the alpha ablation in Appendix D.2 and the headline Meta-World results in Figure 4 both include the same 5 hard tasks (Assembly, Hand Insert, Pick Out of Hole, Pick Place, Push), selection of alpha on those tasks would make the Figure 4 averages partially selected on the evaluation set. Please disclose the search range, the number of seeds used for selection, the selection criterion, and ideally tune alpha on a separate task split or report sensitivity curves for all 15 tasks; the ManiSkill3 transfer of alpha=0.8 is suggestive but does not remove this concern.
  2. [Section 4.3, Eq. (1)-(2), Figure 5 (top)] The ablation labeled 'MAD - Naive Both' does not isolate the SADA stabilization recipe from the alpha reweighting. MAD uses the SADA-structured losses with alpha=0.8, while 'MAD - Naive Both' is described only as training on merged and singular view features 'without using the MAD loss formulations'; no alpha-weighted version of the naive objective is reported. As a result, the comparison cannot distinguish whether the stability comes from the SADA target-selection structure (targets predicted from the unaugmented stream) or simply from the high weight on the unaugmented merged-view loss. Because the paper's central assumption is that the SADA recipe transfers from image-level augmentation to feature-level augmentation, please add an ablation that applies the same alpha weighting to a naive merged-plus-singular objective, or otherwise support the attribution of stability to the SADA recipe.
  3. [Section 5, Baselines; Appendix A.2] Baseline implementations are not specified in enough detail to verify the central comparison. The text says MVD, VIB, and Single Camera DrQ are reimplemented on top of the authors' DrQ baseline, and that MV-MWM is kept as a model-based baseline and 'tuned accordingly', but no hyperparameters, architecture details, loss weights, or training budgets are provided for these reimplementations. In addition, Section 5 says MV-MWM is trained 'without any expert demonstrations for a fair comparison', while Appendix A.2 says the MV-MWM baseline was kept as its model-based baseline; this should be clarified. Please provide full baseline configurations or a code release, because the reported 30% and 36% improvements over baselines are the paper's main empirical claim.
minor comments (5)
  1. [Section 3 and Eq. (2)] The statement 'Qθ represents an exponential moving average of the weights from Qθ' is self-contradictory; please introduce explicit target-network notation (for example, Q-bar) and use it consistently in the critic target in Eq. (2).
  2. [Section 3 and Section 4.3] The symbol alpha is used both for the SAC entropy temperature and for the MAD loss weight; this overloads the notation in Eq. (1)-(2) and Table 1. Please rename one of the two hyperparameters.
  3. [Section 4.1] The sentence 'Feature summation preserves the magnitude of different view features, such that the downstream actor and critic have a signal of how many views are inputted' is unclear, because summation makes the feature magnitude grow with the number of views rather than preserving a fixed magnitude. Please clarify the intended property.
  4. [Section 5 and Appendix A.1] The observation shape is described as '(3×R(3×84×84))' in Section 5 but as 'Image Size (3, 84, 84)' in Table 1; please make the frame-stack and channel dimensions consistent across the two descriptions.
  5. [Figure 4 and Figure 5] The final success values on the far-right 'Average' panels are very small and difficult to read in print; a table with the final mean and CI for each method and evaluation condition would improve reproducibility of the headline numbers.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation; MAD's benchmark claims are externally validated, with only a minor non-load-bearing self-citation to SADA.

full rationale

The derivation chain is self-contained against external benchmarks. MAD's merge module sums per-view features to form Mt, and the loss in Eqs. (1)-(2) is a convex combination of an unaugmented merged-feature objective and single-view feature-augmented objectives. Setting alpha=0.5 recovers the SADA loss, but SADA is prior published work by overlapping authors, not a quantity fitted in this paper, and the reported success rates come from direct evaluation against MVD, VIB, MV-MWM, and Single Camera DrQ on Meta-World and ManiSkill3. The only self-citation is SADA [8], used to justify the selective-augmentation recipe; this is a reliance rather than a hidden fit because the central claim of improved sample efficiency and robustness is demonstrated in Figures 4-8, not derived from SADA by construction. The alpha tuning and ablation are a statistical design concern about possible selection on the evaluation set, not a circularity, since the reported curves are measurements rather than predictions constructed from the fitted alpha. No equation or definition reduces to its own output, so there is no circular step.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central empirical claim is supported by benchmarks, but the method imports its stabilization mechanism from the self-cited SADA paper and assumes that summed features from a shared encoder are a valid state representation for the actor and critic. The only new number fit to data is alpha.

free parameters (1)
  • alpha loss weight = 0.8
    Weighs the merged (unaugmented) stream against the single-view augmented streams in Eqs. (1)-(2). Tuned via ablation on 5 Meta-World hard tasks (Figure 5, bottom); no held-out split is described for the tuning decision.
assumptions (4)
  • domain assumption The SADA selective-augmentation recipe stabilizes actor-critic learning when single-view features are used as augmentations.
    Section 4.3 builds the MAD losses on SADA; no new proof or independent validation is provided for feature-level augmentation.
  • domain assumption A single shared CNN encoder maps all camera views into a common feature space where summation is a meaningful merge and each V_i is a valid input for the actor and critic.
    Section 4.1 assumes feature summation preserves view-count magnitude and equal dimensionality; no analysis justifies that summed features are semantically aligned across views.
  • domain assumption DrQ with random shifts remains a suitable backbone when input observations are merged or singular features.
    The implementation section selects DrQ and applies identical random shifts to all views; no proof is offered that the base algorithm's assumptions hold for the MAD objective.
  • domain assumption Success rate over 20 episodes with 5 seeds is sufficient to compare policies.
    The evaluation protocol in Section 5 reports mean success rate and 95% CIs but no statistical tests across tasks or seeds.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Merging and Disentangling Views in Visual Reinforcement Learning for Robotic Manipulation." pith.science (2026). https://pith.science/paper/AMG4UURQ

@misc{pith2026250504619,
  author       = {Pith},
  title        = {Pith review of: Merging and Disentangling Views in Visual Reinforcement Learning for Robotic Manipulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AMG4UURQ}},
  note         = {Machine review of arXiv:2505.04619}
}
read the original abstract

Vision is well-known for its use in manipulation, especially using visual servoing. Due to the 3D nature of the world, using multiple camera views and merging them creates better representations for Q-learning and in turn, trains more sample efficient policies. Nevertheless, these multi-view policies are sensitive to failing cameras and can be burdensome to deploy. To mitigate these issues, we introduce a Merge And Disentanglement (MAD) algorithm that efficiently merges views to increase sample efficiency while simultaneously disentangling views by augmenting multi-view feature inputs with single-view features. This produces robust policies and allows lightweight deployment. We demonstrate the efficiency and robustness of our approach using Meta-World and ManiSkill3. For project website and code, see https://aalmuzairee.github.io/mad

Figures

Figures reproduced from arXiv: 2505.04619 by the authors.

Figure 1
Figure 1. Merge And Disentangle. We introduce a method that can merge multiple camera views during training to learn better representations, while simultaneously disentangling the camera view representations, such that the policy can function with any singular view input during deployment. Abstract: Vision is well-known for its use in manipulation, especially using vi￾sual servoing. Due to the 3D nature of the world, using mu… view at source ↗
Figure 2
Figure 2. Environment Setup. Our environment setup for robotic manipulation where we use three input camera views as inputs, namely First Person, Third Person A, and Third Person B across two visual RL benchmarks: (Left) Meta-World (Right) ManiSkill3. Extended visuals in Appendix E. robust to a reduction in available camera views, policies must be carefully disentangled during training [7]. Aligning these two directions, we p… view at source ↗
Figure 3
Figure 3. Our framework. Update diagram of a generic visual actor-critic model with our modifi￾cations. MAD merges camera views through feature summation and disentangles camera views by selectively augmenting inputs to the downstream actor and critic with all the singular view features. The agent is trained end-to-end with our defined MAD loss functions. (Left): Single Shared CNN Encoder. (Middle): Actor Update diagram (Righ… view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Overall Robustness. Success rate as a function of environment steps, averaged over all (Top) 15 Meta-World and (Bottom) 5 ManiSkill3 visual RL tasks. Methods are trained on all three camera views and evaluated on all and singular camera views with the final average dis…
Figure 5
Figure 5. Figure 5: Ablations. Success rate as a function of environment steps, averaged over 5 Meta-World hard tasks. (Top) Component Ablations. (Bottom) Alpha Ablations. Methods trained on all camera views and evaluated on all and singular camera views. Mean and 95% CI over 5 random see…
Figure 6
Figure 6. Figure 6: Multi-view Merging. Success rate as a function of environment steps averaged over 5 Meta-World hard visual RL tasks. Methods were trained and evaluated on (Left) two camera views - First and Third A - (Right) three cam￾era views. Dashed line indicates highest per￾forma…
Figure 8
Figure 8. Figure 8: Modality Adaptability. Success rate as a function of environment steps averaged over 5 ManiSkill3 tasks. (Left) RGB and Depth images from the PokeCube Third Person A view. (Right) Methods are trained on one camera view (Third Person A) but with different modalities, an…
Figure 9
Figure 9. Figure 9: View Robustness. Success rate as a function of environment steps, averaged over 5 ManiSkill3 visual RL tasks. Methods are trained on all five camera views and evaluated on all and singular camera views with the final average displayed on the far right. Mean and 95% CI …
Figure 10
Figure 10. Figure 10: View Robustness. Success rate as a function of environment steps, averaged over 5 ManiSkill3 visual RL tasks. Methods are trained on all three camera views and evaluated on all and singular camera views with the final average displayed on the far right. Mean and 95% C…
Figure 11
Figure 11. Figure 11: First Person Camera View Failure. Success rate as a function of environment steps on the ManiSkill3 PokeCube task. Methods are trained on all three views and evaluated an all and singular views separately. Mean and 95% CI over 5 random seeds. 18 [PITH_FULL_IMAGE:figu…
Figure 12
Figure 12. Figure 12: Overall Robustness. Success rate as a function of environment steps, averaged over 15 Meta-World and 5 ManiSkill3 visual RL tasks. (Top Left) Meta-World Medium. (Top Right) Meta-World Hard. (Bottom Left) Meta-World Very Hard. (Bottom Right) ManiSkill3. Methods are tra…
Figure 13
Figure 13. Figure 13: Ablations. Success rate as a function of environment steps, averaged over 5 Meta-World hard visual RL tasks. (Left) Alpha Ablations. (Right) Component Ablations. Methods are trained on all three camera views and evaluated on all and singular camera views. Mean and 95%…
Figure 14
Figure 14. Figure 14: Multi-view Merging. Success rate as a function of environment steps, averaged over 5 Meta-World hard visual RL tasks. Methods trained and evaluated on (Left) two camera views - First and Third A - (Right) three camera views. Mean and 95% CI over 5 random seeds. In ref…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

72 extracted references · 32 canonical work pages

  1. [1]

    V . Mnih, K. Kavukcuoglu, D. Silver, A. Graves, I. Antonoglou, D. Wierstra, and M. Riedmiller. Playing atari with deep reinforcement learning. arXiv preprint arXiv:1312.5602, 2013

    arXiv 2013

  2. [2]

    Levine, C

    S. Levine, C. Finn, T. Darrell, and P. Abbeel. End-to-end training of deep visuomotor policies. The Journal of Machine Learning Research, 17(1):1334–1373, 2016

    2016

  3. [3]

    Pinto and A

    L. Pinto and A. Gupta. Supersizing self-supervision: Learning to grasp from 50k tries and 700 robot hours. In 2016 IEEE international conference on robotics and automation (ICRA), pages 3406–3413. IEEE, 2016

    2016

  4. [4]

    Kalashnikov, A

    D. Kalashnikov, A. Irpan, P. Pastor, J. Ibarz, A. Herzog, E. Jang, D. Quillen, E. Holly, M. Kalakrishnan, V . Vanhoucke, et al. Qt-opt: Scalable deep reinforcement learning for vision- based robotic manipulation. arXiv preprint arXiv:1806.10293, 2018

    arXiv 2018

  5. [5]

    Kaufmann, L

    E. Kaufmann, L. Bauersfeld, A. Loquercio, M. M ¨uller, V . Koltun, and D. Scara- muzza. Champion-level drone racing using deep reinforcement learning. Nat., 620(7976): 982–987, 2023. doi:10.1038/S41586-023-06419-4. URL https://doi.org/10.1038/ s41586-023-06419-4

    doi:10.1038/s41586-023-06419-4 2023

  6. [6]

    K. Hsu, M. J. Kim, R. Rafailov, J. Wu, and C. Finn. Vision-based manipulators need to also see from their hands. arXiv preprint arXiv:2203.12677, 2022

    arXiv 2022

  7. [7]

    Dunion and S

    M. Dunion and S. V . Albrecht. Multi-view disentanglement for reinforcement learning with multiple cameras. arXiv preprint arXiv:2404.14064, 2024

    arXiv 2024

  8. [8]

    Almuzairee, N

    A. Almuzairee, N. Hansen, and H. I. Christensen. A recipe for unbounded data augmentation in visual reinforcement learning. arXiv preprint arXiv:2405.17416, 2024

    arXiv 2024

Show all 72 references
  1. [9]

    T. Yu, D. Quillen, Z. He, R. Julian, K. Hausman, C. Finn, and S. Levine. Meta-world: A benchmark and evaluation for multi-task and meta reinforcement learning. In Conference on robot learning, pages 1094–1100. PMLR, 2020

    2020

  2. [10]

    S. Tao, F. Xiang, A. Shukla, Y . Qin, X. Hinrichsen, X. Yuan, C. Bao, X. Lin, Y . Liu, T.-k. Chan, et al. Maniskill3: Gpu parallelized robotics simulation and rendering for generalizable embodied ai. arXiv preprint arXiv:2410.00425, 2024

    arXiv

  3. [11]

    Noroozi and P

    M. Noroozi and P. Favaro. Unsupervised learning of visual representations by solving jigsaw puzzles. In European Conference on Computer Vision, pages 69–84. Springer, 2016

    2016

  4. [12]

    Z. Wu, Y . Xiong, S. X. Yu, and D. Lin. Unsupervised feature learning via non-parametric instance discrimination. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 3733–3742, 2018

    2018

  5. [13]

    A. v. d. Oord, Y . Li, and O. Vinyals. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748, 2018

    2018 arXiv

  6. [14]

    Y . Tian, D. Krishnan, and P. Isola. Contrastive multiview coding. arXiv preprint arXiv:1906.05849, 2019

    1906 arXiv

  7. [15]

    T. Chen, S. Kornblith, M. Norouzi, and G. Hinton. A simple framework for contrastive learning of visual representations. In International conference on machine learning, pages 1597–1607. PMLR, 2020. 10

    2020

  8. [16]

    K. He, X. Chen, S. Xie, Y . Li, P. Doll ´ar, and R. Girshick. Masked autoencoders are scalable vision learners. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16000–16009, 2022

    2022

  9. [17]

    Srinivas, M

    A. Srinivas, M. Laskin, and P. Abbeel. Curl: Contrastive unsupervised representations for reinforcement learning. arXiv preprint arXiv:2004.04136, 2020

    2004 arXiv

  10. [18]

    Laskin, K

    M. Laskin, K. Lee, A. Stooke, L. Pinto, P. Abbeel, and A. Srinivas. Reinforcement learning with augmented data. arXiv preprint arXiv:2004.14990, 2020

    2004 arXiv

  11. [19]

    Kostrikov, D

    I. Kostrikov, D. Yarats, and R. Fergus. Image augmentation is all you need: Regularizing deep reinforcement learning from pixels. International Conference on Learning Representations , 2020

    2020

  12. [20]

    Stooke, K

    A. Stooke, K. Lee, P. Abbeel, and M. Laskin. Decoupling representation learning from rein- forcement learning. ArXiv, abs/2004.1499, 2020

    2004

  13. [21]

    Yarats, R

    D. Yarats, R. Fergus, A. Lazaric, and L. Pinto. Reinforcement learning with prototypical representations. arXiv preprint arXiv:2102.11271, 2021

    2021 arXiv

  14. [22]

    Yarats, R

    D. Yarats, R. Fergus, A. Lazaric, and L. Pinto. Mastering visual continuous control: Improved data-augmented reinforcement learning. arXiv preprint arXiv:2107.09645, 2021

    2021 arXiv

  15. [23]

    Raileanu, M

    R. Raileanu, M. Goldstein, D. Yarats, I. Kostrikov, and R. Fergus. Automatic data augmenta- tion for generalization in deep reinforcement learning. ArXiv, abs/2006.12862, 2020

    2006 arXiv

  16. [24]

    Hansen, H

    N. Hansen, H. Su, and X. Wang. Stabilizing deep q-learning with convnets and vision trans- formers under data augmentation. In Annual Conference on Neural Information Processing Systems (NeurIPS), 2021

    2021

  17. [25]

    X. Wang, L. Lian, and S. X. Yu. Unsupervised visual attention and invariance for reinforcement learning. ArXiv, abs/2104.02921, 2021

    2021 arXiv

  18. [26]

    L. Fan, G. Wang, D.-A. Huang, Z. Yu, L. Fei-Fei, Y . Zhu, and A. Anandkumar. Secant: Self-expert cloning for zero-shot generalization of visual policies. In M. Meila and T. Zhang, editors, Proceedings of the 38th International Conference on Machine Learning , volume 139 of Pro...

    2021

  19. [27]

    Z. Yuan, G. Ma, Y . Mu, B. Xia, B. Yuan, X. Wang, P. Luo, and H. Xu. Don’t touch what mat- ters: Task-aware lipschitz data augmentation for visual reinforcement learning. arXiv preprint arXiv:2202.09982, 2022

    2022 arXiv

  20. [28]

    Grooten, T

    B. Grooten, T. Tomilin, G. Vasan, M. E. Taylor, A. R. Mahmood, M. Fang, M. Pechenizkiy, and D. C. Mocanu. Madi: Learning to mask distractions for generalization in visual deep reinforcement learning. arXiv preprint arXiv:2312.15339, 2023

    2023 arXiv

  21. [29]

    S. Yang, Y . Ze, and H. Xu. Movie: Visual model-based policy adaptation for view generaliza- tion. Advances in Neural Information Processing Systems, 36, 2024

    2024

  22. [30]

    S. Y . Lee, C. Sungik, and S.-Y . Chung. Sample-efficient deep reinforcement learning via episodic backward update. Advances in neural information processing systems, 32, 2019

    2019

  23. [31]

    K. Wang, B. Kang, J. Shao, and J. Feng. Improving generalization in reinforcement learning with mixture regularization. Advances in Neural Information Processing Systems , 33:7968– 7978, 2020

    2020

  24. [32]

    Hansen and X

    N. Hansen and X. Wang. Generalization in reinforcement learning by soft data augmentation. In International Conference on Robotics and Automation, 2021. 11

    2021

  25. [33]

    Zhang and Y

    H. Zhang and Y . Guo. Generalization of reinforcement learning with policy-aware adversarial data augmentation. arXiv preprint arXiv:2106.15587, 2021

    2021 arXiv

  26. [34]

    Huang, P

    Y . Huang, P. Peng, Y . Zhao, G. Chen, and Y . Tian. Spectrum random masking for generalization in image-based reinforcement learning. Advances in Neural Information Processing Systems, 35:20393–20406, 2022

    2022

  27. [35]

    Z. Wang, Y . Ze, Y . Sun, Z. Yuan, and H. Xu. Generalizable visual reinforcement learning with segment anything model. arXiv preprint arXiv:2312.17116, 2023

    2023 arXiv

  28. [36]

    E. Teoh, S. Patidar, X. Ma, and S. James. Green screen augmentation enables scene generali- sation in robotic manipulation. arXiv preprint arXiv:2407.07868, 2024

    2024 arXiv

  29. [37]

    Akinola, J

    I. Akinola, J. Varley, and D. Kalashnikov. Learning precise 3d manipulation from multiple uncalibrated cameras. In 2020 IEEE International Conference on Robotics and Automation (ICRA), pages 4616–4622. IEEE, 2020

    2020

  30. [38]

    Y . Seo, J. Kim, S. James, K. Lee, J. Shin, and P. Abbeel. Multi-view masked world models for visual robotic manipulation. In International Conference on Machine Learning , pages 30613–30632. PMLR, 2023

    2023

  31. [39]

    Gervet, Z

    T. Gervet, Z. Xian, N. Gkanatsios, and K. Fragkiadaki. Act3d: Infinite resolution action detec- tion transformer for robotic manipulation. arXiv preprint arXiv:2306.17817, 2023

    2023 arXiv

  32. [40]

    Goyal, V

    A. Goyal, V . Blukis, J. Xu, Y . Guo, Y .-W. Chao, and D. Fox. Rvt-2: Learning precise manipu- lation from few demonstrations. arXiv preprint arXiv:2406.08545, 2024

    2024 arXiv

  33. [41]

    S. Qian, K. Mo, V . Blukis, D. F. Fouhey, D. Fox, and A. Goyal. 3d-mvp: 3d multiview pretraining for robotic manipulation. arXiv preprint arXiv:2406.18158, 2024

    2024 arXiv

  34. [42]

    Z. Yuan, T. Wei, S. Cheng, G. Zhang, Y . Chen, and H. Xu. Learning to manipulate anywhere: A visual generalizable framework for reinforcement learning. arXiv preprint arXiv:2407.15815, 2024

    2024 arXiv

  35. [43]

    A. Szot, A. Clegg, E. Undersander, E. Wijmans, Y . Zhao, J. Turner, N. Maestre, M. Mukadam, D. S. Chaplot, O. Maksymets, et al. Habitat 2.0: Training home assistants to rearrange their habitat. Advances in neural information processing systems, 34:251–266, 2021

    2021

  36. [44]

    Jangir, N

    R. Jangir, N. Hansen, S. Ghosal, M. Jain, and X. Wang. Look closer: Bridging egocentric and third-person views with transformers for robotic manipulation.IEEE Robotics and Automation Letters, 7(2):3046–3053, 2022

    2022

  37. [45]

    H. Su, S. Maji, E. Kalogerakis, and E. Learned-Miller. Multi-view convolutional neural net- works for 3d shape recognition. In Proceedings of the IEEE international conference on com- puter vision, pages 945–953, 2015

    2015

  38. [46]

    Y . Yao, Z. Luo, S. Li, T. Fang, and L. Quan. Mvsnet: Depth inference for unstructured multi- view stereo. In Proceedings of the European conference on computer vision (ECCV) , pages 767–783, 2018

    2018

  39. [47]

    Borse, M

    S. Borse, M. Klingner, V . R. Kumar, H. Cai, A. Almuzairee, S. Yogamani, and F. Porikli. X- align: Cross-modal cross-view alignment for bird’s-eye-view segmentation. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 3287– 3297, J...

    2023

  40. [48]

    Z. Li, W. Wang, H. Li, E. Xie, C. Sima, T. Lu, Q. Yu, and J. Dai. Bevformer: Learning bird’s- eye-view representation from lidar-camera via spatiotemporal transformers. IEEE Transac- tions on Pattern Analysis and Machine Intelligence, 2024. 12

    2024

  41. [49]

    Khosla, P

    P. Khosla, P. Teterwak, C. Wang, A. Sarna, Y . Tian, P. Isola, A. Maschinot, C. Liu, and D. Kr- ishnan. Supervised contrastive learning. Advances in neural information processing systems, 33:18661–18673, 2020

    2020

  42. [50]

    Oquab, T

    M. Oquab, T. Darcet, T. Moutakanni, H. V o, M. Szafraniec, V . Khalidov, P. Fernandez, D. Haz- iza, F. Massa, A. El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023

    2023 arXiv

  43. [51]

    M. Li, L. Wu, J. Wang, and H. Bou Ammar. Multi-view reinforcement learning. Advances in neural information processing systems, 32, 2019

    2019

  44. [52]

    H. Yang, D. Shi, G. Xie, Y . Peng, Y . Zhang, Y . Yang, and S. Yang. Self-supervised repre- sentations for multi-view reinforcement learning. In The 38th Conference on Uncertainty in Artificial Intelligence, 2022

    2022

  45. [53]

    Eysenbach, A

    B. Eysenbach, A. Gupta, J. Ibarz, and S. Levine. Diversity is all you need: Learning skills without a reward function. arXiv preprint arXiv:1802.06070, 2018

    2018 arXiv

  46. [54]

    S. Park, O. Rybkin, and S. Levine. Metra: Scalable unsupervised rl with metric-aware abstrac- tion. arXiv preprint arXiv:2310.08887, 2023

    2023 arXiv

  47. [55]

    Vasco, H

    M. Vasco, H. Yin, F. S. Melo, and A. Paiva. How to sense the world: Leveraging hi- erarchy in multimodal perception for robust reinforcement learning agents. arXiv preprint arXiv:2110.03608, 2021

    2021 arXiv

  48. [56]

    E. S. Hu, J. Springer, O. Rybkin, and D. Jayaraman. Privileged sensing scaffolds reinforcement learning. arXiv preprint arXiv:2405.14853, 2024

    2024 arXiv

  49. [57]

    Skand, B

    S. Skand, B. Pandit, C. Kim, L. Fuxin, and S. Lee. Simple masked training strategies yield control policies that are robust to sensor failure. In 8th Annual Conference on Robot Learning,

  50. [58]

    L. P. Kaelbling, M. L. Littman, and A. R. Cassandra. Planning and acting in partially observ- able stochastic domains. Artificial Intelligence, 1998

    1998

  51. [59]

    Haarnoja, A

    T. Haarnoja, A. Zhou, P. Abbeel, and S. Levine. Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor. arXiv preprint arXiv:1801.01290, 2018

    2018 arXiv

  52. [60]

    C. J. Watkins and P. Dayan. Q-learning. Machine learning, 8:279–292, 1992

    1992

  53. [61]

    T. P. Lillicrap, J. J. Hunt, A. Pritzel, N. Heess, T. Erez, Y . Tassa, D. Silver, and D. Wierstra. Continuous control with deep reinforcement learning. arXiv preprint arXiv:1509.02971, 2015

    2015 arXiv

  54. [62]

    V . Mnih, K. Kavukcuoglu, D. Silver, A. A. Rusu, J. Veness, M. G. Bellemare, A. Graves, M. Riedmiller, A. K. Fidjeland, G. Ostrovski, et al. Human-level control through deep rein- forcement learning. nature, 518(7540):529–533, 2015

    2015

  55. [63]

    Y . Seo, D. Hafner, H. Liu, F. Liu, S. James, K. Lee, and P. Abbeel. Masked world models for visual control. In Conference on Robot Learning, pages 1332–1344. PMLR, 2023

    2023

  56. [64]

    Lancaster, N

    P. Lancaster, N. Hansen, A. Rajeswaran, and V . Kumar. Modem-v2: Visuo-motor world models for real-world robot manipulation. In 2024 IEEE International Conference on Robotics and Automation (ICRA), pages 7530–7537. IEEE, 2024

    2024

  57. [65]

    S. Jang, H. Jeong, and H. Yang. Murm: utilization of multi-views for goal-conditioned rein- forcement learning in robotic manipulation. Robotics, 12(4):119, 2023

    2023

  58. [66]

    T. Lin, K. Sachdev, L. Fan, J. Malik, and Y . Zhu. Sim-to-real reinforcement learning for vision-based dexterous manipulation on humanoids. arXiv preprint arXiv:2502.20396, 2025. 13

    2025 arXiv

  59. [67]

    H. Hu, S. Mirchandani, and D. Sadigh. Imitation bootstrapped reinforcement learning. arXiv preprint arXiv:2311.02198, 2023

    2023 arXiv

  60. [68]

    G. Xu, R. Zheng, Y . Liang, X. Wang, Z. Yuan, T. Ji, Y . Luo, X. Liu, J. Yuan, P. Hua, et al. Drm: Mastering visual reinforcement learning through dormant ratio minimization. arXiv preprint arXiv:2310.19668, 2023

    2023 arXiv

  61. [69]

    Hansen, H

    N. Hansen, H. Su, and X. Wang. TD-MPC2: Scalable, robust world models for continuous control. In International Conference on Learning Representations (ICLR), 2024

    2024

  62. [70]

    Fujimoto, P

    S. Fujimoto, P. D’Oro, A. Zhang, Y . Tian, and M. Rabbat. Towards general-purpose model-free reinforcement learning. arXiv preprint arXiv:2501.16142, 2025

    2025 arXiv

  63. [71]

    Van Hoorick, R

    B. Van Hoorick, R. Wu, E. Ozguroglu, K. Sargent, R. Liu, P. Tokmakov, A. Dave, C. Zheng, and C. V ondrick. Generative camera dolly: Extreme monocular dynamic novel view synthesis. In European Conference on Computer Vision, pages 313–331. Springer, 2024. 14 A Experiment Setup A...

    2024

  64. [2024]

    URL https://openreview.net/forum?id=AsbyZRdqPv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.