Pith. sign in

REVIEW 3 major objections 5 minor 48 references

LAWM-3D: Learning 3D-Aware Latent Actions from Human Videos for Generalizable Robot World Models

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 This paper claims that latent action models can be made 3D-aware, and that this is what lets robot world models generalize across scenes.

desk verdict A solid, broadly evaluated integration of multi-view, depth, and geometric alignment for latent actions, but the paper's central leak-prevention claim needs a direct test. read the letter →

arxiv 2608.05706 v1 pith:HL2SDXPL submitted 2026-08-06 cs.CV

Jiarui Yang , Jiale Zhange , Jiawei Li , Hang Guo , Wen Huang , Jinpeng Wang , Peidong Liu , Shu-Tao Xia This is my paper · ORCID
classification cs.CV
keywords latentactionmodelsworld3D-awarerepresentationsmulti-viewvideolearningrobotmanipulationRGB-Dreconstructiongeometricalignmentgeneration
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

This paper argues that simply adding multi-view video to latent action models is not enough to give the learned actions 3D awareness; viewpoint discrepancies and future-frame appearance leakage get in the way. It proposes LAWM-3D, a latent action autoencoder with three coupled designs—multi-view invariant tokenization, geometric alignment to a pretrained 3D foundation model, and a non-injective RGB-D reconstruction objective—that together force the latent action to encode view-invariant, geometrically meaningful motion. Trained on large-scale human videos and fine-tuned on robot data, the resulting 3D-aware latent actions improve a downstream diffusion world model in generation quality, physical consistency, controllability, and out-of-distribution generalization. If correct, this gives a route to robot world models that learn motion semantics from unlabeled video rather than from expensive action annotations.

What carries the argument

The central object is the 3D-aware latent action $a$, a 32-dimensional vector produced by a multi-view spatio-temporal Transformer encoder from a set of synchronized observations $F_{t:t+1}$. Three coupled mechanisms carry the argument: (1) multi-view invariant tokenization, which randomly masks views during training and aggregates the action tokens by mean pooling so the same motion produces the same action from any viewpoint; (2) geometric representation alignment, a cosine and scale loss that anchors the encoder's intermediate layers ($h=[6,11]$) to the frozen VGGT foundation model's view-consistent 3D features; and (3) a non-injective RGB-D joint reconstruction objective, where the decoder predicts both future frame and depth from current RGB-D input, with depth computed offline and never given to the encoder, so the latent action cannot cheat by copying future appearance. The claim is that these together make $a$ a pure, geometrically grounded motion code that a diffusion world model can condition on.

What would settle it

Give the trained encoder only the current frame $f_t$ (with $f_{t+1}$ withheld) and measure how well the world model still reconstructs the future; or train a linear probe to decode $f_{t+1}$'s appearance from $a$ alone. If future texture survives in $a$, the leakage-prevention claim is false and the attribution of downstream gains to 3D awareness is unsupported.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a latent action representation learned from unlabeled video can be made 3D-aware, and that this is what unlocks world-model gains. Naive multi-view training fails because the encoder can copy future-frame RGB content into the latent action and because different cameras disagree on appearance. The paper's three mechanisms close exactly those holes: random view masking plus cross-view aggregation makes the action token view-invariant; aligning intermediate encoder features with VGGT injects explicit cross-view geometric correspondences; and decoding jointly into RGB and depth—with depth unavailable to the encoder—removes the appearance shortcut. Through a two-stage human-video pretraining and robot post-training pipeline, LAWM-3D reports state-of-the-art results across 16 benchmark metrics and on out-of-distribution rollouts, with the largest gains in motion quality and 3D accuracy.

Load-bearing premise

The load-bearing assumption is that the non-injective RGB-D objective actually stops the encoder from copying the future frame's appearance into the latent action, so that the measured downstream gains are caused by 3D awareness rather than by information still leaking into $a$.

Editorial extensions

If this is right

  • If latent actions are truly 3D-aware and view-invariant, a world model conditioned on them should produce rollouts that stay physically consistent longer, since the action code tracks 3D motion rather than 2D pixels.
  • The two-stage recipe (human-video pretraining plus robot post-training) becomes more data-efficient: unlabeled multi-view human videos can substitute for expensive robot action annotations in world-model pretraining.
  • Because the latent action model is frozen and not needed at rollout time, the 3D awareness comes at negligible inference cost, so the approach can be dropped into existing world-model backbones.
  • The learned action space should transfer across viewpoints and embodiments, meaning a single action vector extracted from one camera can drive prediction from another camera's observation.
  • Out-of-distribution generalization in novel scenes and interactions should be stronger than for 2D latent action models, since geometry alignment anchors features to scene structure rather than appearance.

Reading between the lines

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

  • A direct leakage probe—decoding future texture from $a$ alone with the current frame fixed—would turn the paper's leakage-prevention claim into a measurable quantity; the paper does not report such a probe, but this test is a natural extension of its analysis.
  • The alignment target need not be VGGT specifically; the paper's own ablations show DUSt3R and Depth-Anything-V2 also help, so the mechanism is geometric priors in general. One could extend the approach to single-view data by generating pseudo-multi-view pairs from a depth-based renderer, which would scale pretraining beyond synchronized multi-view corpora.
  • Because the latent action is view-invariant, the same learned action space could serve as a shared control interface across camera rigs in real deployment, letting a policy condition on actions extracted from whatever camera is available.
  • The world-model-based policy selection experiments suggest imagined returns correlate strongly with real task performance (Spearman $\rho=0.952$); an untested but plausible consequence is that the same selection mechanism could rank policies in the wild where no reward signal is available.
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 LAWM-3D, a latent action model trained on multi-view and monocular human videos with three coupled components: multi-view invariant action tokenization, geometric feature alignment to a pretrained 3D foundation model (VGGT), and an RGB-D joint reconstruction objective claimed to prevent future-frame appearance leakage. The learned latent actions are used to condition a Cosmos-Predict2.5 world model that is pretrained on human videos and post-trained on robot data. Experiments compare latent action models on Ego-Exo4D and Assembly101, evaluate world model video generation on WorldArena across 16 metrics, test OOD generalization on three benchmarks, and include ablations of the proposed components plus appendix analyses of view invariance, mutual information, linear probing, and computational cost.

Significance. If the causal attribution holds, this is a meaningful step toward 3D-aware latent action learning and could improve world models for robotics by exploiting abundant unlabeled human videos. The paper's strengths include broad evaluation across many metrics and OOD benchmarks, careful ablated study of the proposed components and loss weights, analysis of layer-range and alignment-target choices, and explicit computational cost accounting. The appendix's mutual-information estimates with three estimators and linear-probing results across ID/OOD tasks provide useful evidence that the latent space contains action-related structure, and the cross-view transfer and cosine-similarity analyses support view-invariance claims. However, the central claim that the proposed objective prevents future-frame leakage and thereby causes the observed gains is not measured, and several evaluation details, especially the reference for depth metrics, need clarification before the results can be fully interpreted.

major comments (3)
  1. [Eq. (2), Section 'Pixel-Agnostic Learning via Depth Map'] The claim that the non-injective RGB-D objective prevents future-frame appearance leakage is not supported by the stated mechanism. In Eq. (2), the encoder remains q_phi(a|f^v_{t:t+1}), so the future frame is an explicit encoder input; with beta = 1e-6 the KL term imposes essentially no information bottleneck; the objective still contains log p(f_{t+1}|a,f_t); and the depth target d_{t+1} is a deterministic function of f_{t+1} in the training data, meaning a latent code that simply copies f_{t+1} also satisfies the depth prediction term. The term 'non-injective' describes the RGB-to-depth mapping, not the latent code. No experiment in the paper measures I(a; f_{t+1} | f_t) or an equivalent leakage diagnostic; the Appendix B mutual-information and linear-probing analyses show that a contains action-related information, but action relevance and appearance leakage are not mutually exclusive. Without such a diagnostic, the causal attribution in the abstract and conclusion that 3D-aware latent actions drive the world-model gains remains underdetermined.
  2. [Table 1 and 'Comparison of LAMs'] Table 1 lists MVP-LAM in the comparison but does not include any MVP-LAM row, so the claimed superiority over this baseline is not empirically documented. In addition, no table reports error bars, confidence intervals, or significance tests, although the text repeatedly uses 'significantly' (e.g., Abstract and the Comparison of LAMs section). Several margins in Table 1 are small (RGB PSNR 33.38 vs 32.56 against DreamDojo-D), and without variance information the stability of the empirical claim cannot be assessed. Please add the missing baseline or justify its exclusion, and report statistical significance or variance estimates for the central comparisons.
  3. [Table 1, Table 2, and 'Limitations'] The depth metrics are not referenced to a defined ground truth. The LAM is trained with Depth-Anything-V2 pseudo-depth (Implementation Details and Limitations), and if the depth PSNR/SSIM in Table 1 and DepthAcc in Table 2 are computed against the same pseudo-depth, those numbers measure fit to the training label generator, not metric 3D accuracy. This matters because the Limitations section itself states that the depth signals are 'inherently noisy' and provide 'relative rather than metric depth,' while the main text uses the depth results to claim stronger 3D scene understanding. Please specify the reference for each depth metric, define DepthAcc, and if pseudo-depth is used, either add a metric-depth or multi-view-consistency evaluation or temper the 3D-accuracy claims.
minor comments (5)
  1. [Eqs. (3)-(4)] The notation in Eqs. (3)-(4) is inconsistent with the stated dimensions: y is defined as K x V x P' x D' but the losses sum over p = 1..P; please clarify how VGGT features are aligned to the encoder patch grid and how P' and D' are handled by the projection head.
  2. [Table 2] Several column names in Table 2 (JEPASimilarity, DynamicDegree, FlowScore, DepthAcc, Perspectivity) are not defined in the text, and it is not stated whether higher is better for each; please add a metric appendix or table footnote.
  3. [Eq. (6)] The velocity v_i in Eq. (6) is not defined; please specify how velocity is computed from the latent video sequence in the temporal consistency loss.
  4. [References and author list] The author list contains a typo ('Jiale Zhange'), and the reference list duplicates Ye et al. as both 2025a and 2025b with identical titles; please correct these.
  5. [Method vs Limitations] The method section states that depth provides 'view-consistent geometric cues,' while the Limitations section notes that the offline depth is noisy, relative, and possibly inconsistent across frames and views; please reconcile these statements.

Circularity Check

0 steps flagged · score 2.0 of 10

No established circularity: the central world-model gains are benchmarked externally, while the depth-metric reference and the leakage-prevention mechanism remain unspecified risks.

full rationale

The paper's load-bearing claim—that 3D-aware latent actions improve world-model generation quality, physical consistency, and generalization—is supported by external benchmarks (WorldArena in Table 2, OOD generalization in Table 3, and the policy-selection studies in Appendix A), which are not equivalent to the LAM training objective. The latent action model is trained with Eqs. (2)-(5), and the downstream world model is a separate Cosmos-Predict2.5 backbone conditioned on the learned latent actions. This keeps the main derivation chain independent of its inputs. Two risks are flagged, but neither is an established circular reduction. First, Table 1 reports depth PSNR/SSIM without specifying the reference depth; the Limitations state that 'the depth signals used for the co-reconstruction objective are pre-computed offline by a monocular depth estimator (Depth-Anything-V2).' If the evaluation reference is the same pseudo-depth used as supervision, the depth numbers measure fit to training labels, but the paper does not say so, so this is a missing specification rather than a demonstrated circularity. Second, the paper asserts that the 'non-injective RGB-D joint reconstruction objective' prevents future-frame appearance leakage, but Eq. (2) still feeds f_{t:t+1} to the encoder and retains the RGB reconstruction term, so the claim is not implied by construction; no I(a; f_{t+1} | f_t) or equivalent leakage diagnostic is reported. This is an unsupported mechanism, not a circular derivation. No self-citation chain or imported uniqueness theorem is load-bearing, so the circularity score remains low.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim rests on several domain assumptions: viewpoint-invariant action semantics, VGGT as a reliable geometric teacher, Depth-Anything-V2 pseudo-depth as valid supervision, and human-to-robot transfer. No new physical entities are introduced. The main free parameters are architectural and loss weights, several of which are selected using the same benchmarks used for evaluation.

free parameters (6)
  • Latent action dimension = 32
    Chosen by hand; the capacity of the learned action code shapes all downstream world-model behavior.
  • VAE KL weight beta = 1e-6
    Follows DreamDojo; at this value the KL bottleneck is very weak, which is directly relevant to the future-frame leakage concern.
  • Angular alignment loss weight lambda_Angular = 0.5
    Empirically set; Table 7 shows LAM PSNR varies from 30.18 to 30.56 as this weight changes.
  • Scale alignment loss weight lambda_Scale = 0.05
    Empirically set; Table 7 shows larger values destabilize training.
  • Aligned encoder layer range h = [6,11]
    Selected by ablation in Table 6; different ranges change PSNR by about 1.2 dB.
  • World model temporal consistency weight lambda_wm = 0.1
    Used in Eq. 6; no ablation or sensitivity analysis is reported for this weight.
assumptions (4)
  • domain assumption Actions in 3D space are invariant across viewpoints.
    Stated in 'Joint Training with Single and Multi-View Data'; the entire cross-view pooling objective presupposes this invariance.
  • domain assumption VGGT's frozen features provide trustworthy cross-view geometric correspondences for the training scenes.
    The geometric alignment losses in Eqs. 3 and 4 assume VGGT is a reliable teacher; the authors themselves note the representation quality is upper-bounded by VGGT.
  • domain assumption Depth maps produced by Depth-Anything-V2 are a valid geometric supervision signal.
    These pseudo-depth maps are used to compute d_{t+1} in the joint RGB-D objective; the authors acknowledge the depth is noisy and only relative.
  • domain assumption Latent actions learned from human videos transfer to robot control actions after post-training.
    The two-stage pretrain/fine-tune paradigm assumes shared action semantics between human and robot embodiments; this premise is inherited from DreamDojo.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LAWM-3D: Learning 3D-Aware Latent Actions from Human Videos for Generalizable Robot World Models." pith.science (2026). https://pith.science/paper/HL2SDXPL

@misc{pith2026260805706,
  author       = {Pith},
  title        = {Pith review of: LAWM-3D: Learning 3D-Aware Latent Actions from Human Videos for Generalizable Robot World Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HL2SDXPL}},
  note         = {Machine review of arXiv:2608.05706}
}
read the original abstract

World models enable agents to perform forward rollout and planning without real-world interaction. However, their application in open-world embodied intelligence remains limited by the high cost of action annotations and the heterogeneity of action spaces across platforms. Recently, latent action models (LAMs) have alleviated this bottleneck by learning action representations directly from unlabeled human videos in a self-supervised manner. Nevertheless, most existing LAMs rely on single-view inputs and operate primarily in 2D pixel space, raising a fundamental question: can simply incorporating multi-view videos into LAM training endow the learned latent actions with 3D-aware perception? Our study shows that the answer is negative. The primary reasons lie in future-frame appearance leakage as well as inter-camera appearance discrepancies and viewpoint variations. To address these issues, we propose LAWM-3D, which introduces three tightly coupled key designs: (1) a multi-view invariant unified action tokenization scheme for learning 3D-aware latent actions; (2) a geometric alignment constraint that anchors intermediate encoder features to a pretrained 3D foundation model, thereby explicitly providing cross-view geometric correspondences; and (3) a non-injective RGB-D joint reconstruction objective that prevents shortcut learning from future-frame appearance information, forcing the LAM to focus supervision on motion cues with geometric significance. Importantly, these components are not simply stacked but are tightly coupled through a unified motivation. Built upon a two-stage paradigm of large-scale human video pretraining followed by robot fine-tuning, extensive experiments demonstrate that the proposed 3D-aware latent actions significantly improve world model performance, achieving SOTA results in generation quality, physical consistency, and generalization ability.

Figures

Figures reproduced from arXiv: 2608.05706 by the authors.

Figure 1
Figure 1. Overview of LAWM-3D framework. The model learns 3D-aware latent actions through joint training on multi-view and single-view datasets. The learned latent action space enables world model pretraining and supports downstream tasks, achieving superior overall performance compared to the baseline method DreamDojo (Gao et al. 2026a). tion space alignment. Extensive qualitative and quantitative experiments show that LAWM-… view at source ↗
Figure 2
Figure 2. Architecture of the 3D-aware latent action model and world model training pipeline. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Visual comparison of depth prediction on ego-view. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Qualitative visualization of world model rollouts [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Comparison of real-world rollouts between ground truth (GT), DreamDojo, and our method on manipulation scenarios. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Impact of different pretraining strategies on world [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Agreement between imagined returns from the [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Closed-loop success rates of different policy se [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: Comparison of long real-world rollouts between ground truth (GT), DreamDojo, and LAWM-3D on manipulation [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]
Figure 10
Figure 10. Figure 10: Cross-view transfer of latent actions, showing that the learned action representation captures view-independent [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: Cosine similarity matrix of latent representations [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]
Figure 12
Figure 12. Figure 12: Mutual information analysis of learned latent ac [PITH_FULL_IMAGE:figures/full_fig_p014_12.png]
Figure 13
Figure 13. Figure 13: Linear probing evaluation of latent action repre [PITH_FULL_IMAGE:figures/full_fig_p014_13.png]
Figure 14
Figure 14. Figure 14: Visualization of VGGT point cloud reconstruction of synchronized multi-view human manipulation videos. [PITH_FULL_IMAGE:figures/full_fig_p016_14.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 11 canonical work pages

  1. [1]

    arXiv preprint arXiv:2602.06949 , year=

    DreamDojo: A Generalist Robot World Model from Large-Scale Human Videos , author=. arXiv preprint arXiv:2602.06949 , year=

    arXiv

  2. [2]

    arXiv preprint arXiv:1803.10122 , volume=

    World models , author=. arXiv preprint arXiv:1803.10122 , volume=

    arXiv

  3. [3]

    arXiv preprint arXiv:2602.08971 , year=

    WorldArena: A Unified Benchmark for Evaluating Perception and Functional Utility of Embodied World Models , author=. arXiv preprint arXiv:2602.08971 , year=

    arXiv

  4. [4]

    Forty-second International Conference on Machine Learning , year=

    AdaWorld: Learning Adaptable World Models with Latent Actions , author=. Forty-second International Conference on Machine Learning , year=

    work page

  5. [5]

    arXiv preprint arXiv:2509.18428 , year=

    Latent action pretraining through world modeling , author=. arXiv preprint arXiv:2509.18428 , year=

    arXiv

  6. [6]

    International Conference on Learning Representations , volume=

    Latent action pretraining from videos , author=. International Conference on Learning Representations , volume=

    work page

  7. [7]

    ACM Computing Surveys , volume=

    Understanding world or predicting future? a comprehensive survey of world models , author=. ACM Computing Surveys , volume=. 2025 , publisher=

    2025

  8. [8]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Driveworld: 4d pre-trained scene understanding via world models for autonomous driving , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

Show all 48 references
  1. [9]

    arXiv preprint arXiv:2505.11709 , year=

    Egodex: Learning dexterous manipulation from large-scale egocentric video , author=. arXiv preprint arXiv:2505.11709 , year=

    arXiv

  2. [10]

    Proceedings of the IEEE international conference on computer vision workshops , pages=

    Snapnet-r: Consistent 3d multi-view semantic labeling for robotics , author=. Proceedings of the IEEE international conference on computer vision workshops , pages=

  3. [11]

    Information , volume=

    From Pixels to Motion: A Systematic Analysis of Translation-Based Video Synthesis Techniques , author=. Information , volume=. 2025 , publisher=

    2025

  4. [12]

    IEEE Transactions on Circuits and Systems for Video Technology , year=

    Learning multi-view stereo with geometry-aware prior , author=. IEEE Transactions on Circuits and Systems for Video Technology , year=

  5. [13]

    Forty-first International Conference on Machine Learning , year=

    Genie: Generative interactive environments , author=. Forty-first International Conference on Machine Learning , year=

  6. [14]

    arXiv preprint arXiv:2505.17006 , year=

    Como: Learning continuous latent motion from internet videos for scalable robot learning , author=. arXiv preprint arXiv:2505.17006 , year=

    arXiv

  7. [15]

    arXiv preprint arXiv:2602.20231 , year=

    UniLACT: Depth-Aware RGB Latent Action Learning for Vision-Language-Action Models , author=. arXiv preprint arXiv:2602.20231 , year=

    arXiv

  8. [16]

    arXiv preprint arXiv:2602.03668 , year=

    MVP-LAM: Learning Action-Centric Latent Action via Cross-Viewpoint Reconstruction , author=. arXiv preprint arXiv:2602.03668 , year=

    arXiv

  9. [17]

    arXiv preprint arXiv:2508.08086 , year=

    Matrix-3d: Omnidirectional explorable 3d world generation , author=. arXiv preprint arXiv:2508.08086 , year=

    arXiv

  10. [18]

    arXiv preprint arXiv:2603.02049 , year=

    WorldStereo: Bridging Camera-Guided Video Generation and Scene Reconstruction via 3D Geometric Memories , author=. arXiv preprint arXiv:2603.02049 , year=

  11. [19]

    The Fourteenth International Conference on Learning Representations , year=

    Geometry Forcing: Marrying Video Diffusion and 3D Representation for Consistent World Modeling , author=. The Fourteenth International Conference on Learning Representations , year=

  12. [20]

    The Fourteenth International Conference on Learning Representations , year=

    FantasyWorld: Geometry-Consistent World Modeling via Unified Video and 3D Prediction , author=. The Fourteenth International Conference on Learning Representations , year=

  13. [21]

    arXiv preprint arXiv:2504.20995 , year=

    Tesseract: learning 4d embodied world models , author=. arXiv preprint arXiv:2504.20995 , year=

    arXiv

  14. [22]

    arXiv preprint arXiv:2603.16669 , year=

    Kinema4D: Kinematic 4D World Modeling for Spatiotemporal Embodied Simulation , author=. arXiv preprint arXiv:2603.16669 , year=

  15. [23]

    European Conference on Computer Vision , pages=

    Rotary position embedding for vision transformer , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  16. [24]

    arXiv preprint arXiv:1312.6114 , year=

    Auto-encoding variational bayes , author=. arXiv preprint arXiv:1312.6114 , year=

    arXiv

  17. [25]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Vggt: Visual geometry grounded transformer , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  18. [26]

    Advances in neural information processing systems , volume=

    Denoising diffusion probabilistic models , author=. Advances in neural information processing systems , volume=

  19. [27]

    arXiv preprint arXiv:2511.00062 , year=

    World simulation with video foundation models for physical ai , author=. arXiv preprint arXiv:2511.00062 , year=

    arXiv

  20. [28]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Scalable diffusion models with transformers , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  21. [29]

    arXiv preprint arXiv:1711.05101 , year=

    Decoupled weight decay regularization , author=. arXiv preprint arXiv:1711.05101 , year=

    arXiv

  22. [30]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Ego-exo4d: Understanding skilled human activity from first-and third-person perspectives , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  23. [31]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Assembly101: A large-scale multi-view video dataset for understanding procedural activities , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  24. [32]

    arXiv preprint arXiv:2503.06669 , year=

    Agibot world colosseo: A large-scale manipulation platform for scalable and intelligent embodied systems , author=. arXiv preprint arXiv:2503.06669 , year=

    arXiv

  25. [33]

    Conference on Robot Learning , pages=

    Learning language-conditioned robot behavior from offline data and crowd-sourced annotation , author=. Conference on Robot Learning , pages=. 2022 , organization=

    2022

  26. [34]

    arXiv preprint arXiv:2212.06817 , year=

    Rt-1: Robotics transformer for real-world control at scale , author=. arXiv preprint arXiv:2212.06817 , year=

    arXiv

  27. [35]

    arXiv preprint arXiv:2403.12945 , year=

    Droid: A large-scale in-the-wild robot manipulation dataset , author=. arXiv preprint arXiv:2403.12945 , year=

    arXiv

  28. [36]

    arXiv preprint arXiv:2511.19861 , year=

    Gigaworld-0: World models as data engine to empower embodied ai , author=. arXiv preprint arXiv:2511.19861 , year=

  29. [37]

    arXiv preprint arXiv:2508.05635 , year=

    Genie envisioner: A unified world foundation platform for robotic manipulation , author=. arXiv preprint arXiv:2508.05635 , year=

    arXiv

  30. [38]

    arXiv preprint arXiv:2509.22642 , year=

    Wow: Towards a world omniscient world model through embodied interaction , author=. arXiv preprint arXiv:2509.22642 , year=

  31. [39]

    2012 IEEE/RSJ international conference on intelligent robots and systems , pages=

    Mujoco: A physics engine for model-based control , author=. 2012 IEEE/RSJ international conference on intelligent robots and systems , pages=. 2012 , organization=

    2012

  32. [40]

    Advances in Neural Information Processing Systems , volume=

    Libero: Benchmarking knowledge transfer for lifelong robot learning , author=. Advances in Neural Information Processing Systems , volume=

  33. [41]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Dust3r: Geometric 3d vision made easy , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  34. [42]

    Advances in Neural Information Processing Systems , volume=

    Depth anything v2 , author=. Advances in Neural Information Processing Systems , volume=

  35. [43]

    arXiv preprint arXiv:2304.07193 , year=

    Dinov2: Learning robust visual features without supervision , author=. arXiv preprint arXiv:2304.07193 , year=

    arXiv

  36. [44]

    arXiv preprint arXiv:1610.01644 , year=

    Understanding intermediate layers using linear classifier probes , author=. arXiv preprint arXiv:1610.01644 , year=

    arXiv

  37. [45]

    Conference on Robot Learning , pages=

    Bridgedata v2: A dataset for robot learning at scale , author=. Conference on Robot Learning , pages=. 2023 , organization=

    2023

  38. [46]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Irasim: A fine-grained world model for robot manipulation , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  39. [47]

    arXiv preprint arXiv:2506.01943 , year=

    Learning video generation for robotic manipulation with collaborative trajectory control , author=. arXiv preprint arXiv:2506.01943 , year=

  40. [48]

    International conference on machine learning , pages=

    Mutual information neural estimation , author=. International conference on machine learning , pages=. 2018 , organization=

    2018

Pith tools

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