Pith. sign in

REVIEW 3 major objections 4 minor 42 references

Latent World Models with Monotone Planning Costs for Image-Goal Navigation

T0 review · 3 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A latent world model trained to rank action sequences by monotone cost matches or outperforms reactive policies on image-goal navigation and transfers zero-shot to a physical robot.

desk verdict MCR's ranking signal and the evaluation metrics both measure deviation from the ground-truth trajectory, so the claim of matching reactive policies 'without imitation' is not supported. read the letter →

arxiv 2608.09073 v1 pith:VWQ25VM3 submitted 2026-08-10 cs.RO

Amirhosein Chahe , Siwei Cai , Lifeng Zhou This is my paper · ORCID
classification cs.RO
keywords image-goalnavigationlatentworldmodelsmonotonecostrankingcross-entropymethodplanningautoregressiverollouttrainingfrozenDINOencoderGNMdatasetzero-shotrobotdeployment
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 paper claims that a latent world model for image-goal navigation fails not from inaccurate prediction alone but from a planning cost that does not reliably rank candidate action sequences. It proposes a JEPA-style model built from a frozen DINO-family encoder and a trainable action-conditioned predictor, trained with an autoregressive rollout loss and a Monotone Cost Ranking (MCR) loss. The MCR loss directly shapes the cosine distance between predicted and goal embeddings so that action sequences deviating further from the recorded trajectory incur higher cost. On the GNM dataset the model reduces orientation error by 2.7× relative to the same-encoder DINO-WM baseline and matches or exceeds reactive policies on three of four metrics, all in latent feature space. The paper also reports zero-shot transfer to a physical Husky robot in five trials.

What carries the argument

The Monotone Cost Ranking (MCR) loss is the central mechanism: a pairwise margin-ranking objective over perturbed action sequences whose margin is proportional to the gap in dead-reckoned trajectory deviation from ground truth. It is supported by an autoregressive rollout loss with a counter-curriculum, which trains the predictor on its own previous outputs just as CEM rollout does at inference. Together they shape the planning cost—cosine distance between the pooled predicted final embedding and the pooled goal embedding—to be monotone near the goal.

What would settle it

A direct test: take a scene with two routes to the same goal that diverge and reconverge; if MCR-trained planning consistently avoids the route with larger dead-reckoned deviation even when that route reaches the goal with equal final proximity, the monotone-cost assumption is falsified.

Watch

Extended reading notes

Core claim

The central finding is that a latent world model can match or exceed reactive policies, without imitation and entirely in latent feature space, if its planning cost is made monotone in trajectory deviation. The paper achieves this with a pairwise margin-ranking loss over graded action perturbations, using the dead-reckoned mean L2 deviation from the ground-truth trajectory as the ranking signal. It also finds that order-sensitive InfoNCE action-contrastive training with temporal permutation negatives distorts the latent geometry and degrades cross-entropy-method planning, a negative result that contrasts with the gains from rollout and MCR training.

Load-bearing premise

The load-bearing premise is that the recorded ground-truth trajectory is a valid monotone reference for goal-reaching: the MCR loss ranks candidate action sequences by their dead-reckoned mean L2 deviation from that trajectory, so any equal-or-better path that deviates from the recorded route will be assigned a higher cost.

Editorial extensions

If this is right

  • Image-goal navigation can be performed entirely in the latent space of a frozen encoder with a trainable predictor, without pixel reconstruction, and with CEM planning.
  • Autoregressive rollout supervision is a direct remedy for teacher-forcing exposure bias in latent world models; the paper's ablations show it lowers AOE from 12.10° to 9.12° on DINOv2 before MCR.
  • Monotone cost ranking improves orientation error over every base variant and both encoders, indicating that cost-shaping is a necessary component for sampling-based MPC.
  • Order-sensitive action-contrastive training should be avoided or redesigned for planning-oriented world models, since temporal permutation negatives erased the gains of rollout training.
  • A latent world model trained only on offline RGB data can transfer zero-shot to a physical Husky in static indoor and outdoor scenes, reaching the goal in five of five limited trials versus two of five for a reactive baseline.

Reading between the lines

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

  • If the monotone-cost principle holds beyond navigation, then any latent world model used with a sampling-based optimizer should train its cost surface, not just its predictions; manipulation and driving are obvious places to test this.
  • A testable variation is to replace the dead-reckoned GT-deviation ranking signal with a goal-proximity measure learned in embedding space, which would remove the reliance on recorded trajectories and could handle multi-path scenes.
  • The contrastive negative result points to a principled fix: separate the contrastive and predictive objectives with a projection head or detached branch, then measure whether CEM planning recovers on the GNM benchmark.
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 / 4 minor

Summary. The paper proposes a latent world model for image-goal navigation built on a frozen DINO-family encoder and a trainable AdaLN ViT predictor. The training procedure combines an autoregressive rollout loss with a Monotone Cost Ranking (MCR) loss that shapes the cosine planning cost so that action sequences with larger dead-reckoned deviation from the ground-truth trajectory receive higher cost. Planning is performed with CEM in latent space. On the GNM dataset, the method is reported to outperform NWM, DINO-WM, OmniVLA, and NoMaD on several metrics, and a zero-shot deployment on a physical Husky is described. The paper also reports a negative result for order-sensitive action-contrastive training.

Significance. If the central claim is established, the contributions are useful and timely: the rollout-loss counter-curriculum is a sensible fix for the teacher-forcing/MPC mismatch, and the MCR objective directly addresses a real problem for latent-space MPC, namely that a predictive model with accurate features can still induce a poorly ordered planning cost. The ablation table is informative and the negative result on order-sensitive contrastive learning is a valuable caution for the world-model community. The real-robot deployment, though small, supports feasibility. However, the paper's headline claim — that a latent world model matches or exceeds reactive policies "without imitation" — is not currently established by the reported metrics, because the training signal and the evaluation metrics both measure agreement with the recorded ground-truth trajectory rather than goal-reaching itself.

major comments (3)
  1. [§4.3, §4.4, Eq. (6)] All four metrics in Table 1 (AOE/MAOE/ADE/MADE) measure deviation from the recorded ground-truth trajectory, and the MCR training signal in Eq. (6) is exactly the same dead-reckoned deviation d_i. The central claim in §4.4 — that the model matches reactive policies "without imitation" and "is not penalized for reaching the goal via a different path" — is therefore not supported by these metrics: ADE/MADE penalize any different path, and MCR is explicitly shaped toward the recorded trajectory. The paper needs a goal-only success metric (for example, success rate based on whether the final reached observation is visually consistent with the goal image, or final metric distance to the goal location) to separate trajectory-following from goal-reaching.
  2. [§3.3.2] The MCR loss assumes that larger dead-reckoned deviation from the recorded ground-truth trajectory implies worse goal-reaching. In the GNM cross-embodiment dataset, a goal image can often be reached by multiple valid routes, so this assumption can assign higher cost to valid alternative trajectories and thereby shape the planning landscape away from goal-proximity. The authors should test this structural premise directly, for example by comparing CEM plans under MCR-shaped costs with plans under a goal-proximity-only cost on a goal-based success measure.
  3. [Table 1, §4.2] Table 1 reports no error bars, no multiple-seed results, and no significance tests. Several headline comparisons are small (e.g., MCR(DINOv2) AOE 7.63 vs. OmniVLA 7.73), and the 2.7× reduction relative to DINO-WM is measured against the authors' internal reimplementation, not against the published DINO-WM numbers. Please report variance over seeds and validate the DINO-WM reimplementation against its published results, or the state-of-the-art claim is not quantitatively robust.
minor comments (4)
  1. [§4.5] The text states that "MCR gives the largest and most consistent per-stage improvement," but Table 1 reports only final metrics; a per-stage table or explicit per-stage comparisons would make this claim directly testable.
  2. [Appendix B.2, Table 3] The counter-curriculum schedule lists batch sizes as "per GPU" in the caption and text, but Table 2 describes "Batch/GPU" with the same values; please make the per-GPU vs. total batch-size convention consistent throughout.
  3. [Eq. (3), Table 2] The per-step prediction loss uses λ_cos = 0.5, but Table 2 does not list λ_cos among the shared hyperparameters; include it for reproducibility.
  4. [Figure 3, §4.6] The method is referred to as "WorldModel" in Figure 3 and the deployment section, while the rest of the paper uses "ours" or "MCR"; please use one consistent name.

Circularity Check

1 steps flagged · score 1.0 of 10

No definitional or self-citation circularity; the MCR trajectory-deviation signal and the GT-based evaluation metrics overlap, giving a minor construct-level caveat to the 'without imitation' claim.

  1. other [Section 3.3.2 (Eq. 6) and Section 4.3 (Metrics); cf. Section 4.4]
    "For each candidate i we compute (1) its cosine cost ci = 1 − cos(F H θ ( ˜Zctx,a (i) 1:H ), ˜zg) and (2) its trajectory distance di (mean L2 deviation of the dead-reckoned path from the GT trajectory). ... Over the K= 6 step trajectory we report orientation error (AOE/MAOE, °; average and worst-case heading error, following 13) and displacement error (ADE/MADE, m; average and worst-case L2 distance from GT)."

    The MCR ranking signal d_i and the headline displacement metrics ADE/MADE are both L2 deviations from the same recorded GT trajectories, so the reported low ADE/MADE values partially reward reproducing training-distribution paths rather than independently reaching image goals. The paper's statement in Section 4.4 that the model 'is not penalized for reaching the goal via a different path' is inconsistent with ADE/MADE, which penalize any different path. However, MCR optimizes a latent cosine-cost ordering, not AOE/ADE directly, and evaluation is on a held-out split against external baselines, so this is a construct-overlap weakness rather than a definitional reduction.

full rationale

The paper's core derivation is self-contained: the autoregressive rollout loss (Eq. 5) supervises the predictor on its own outputs, MCR (Eq. 6) shapes the cosine cost using dead-reckoned trajectory deviations, and CEM (Eq. 8) plans by minimizing that cost in latent space. No parameter is fitted to the test metrics, no load-bearing claim rests on a self-citation (the authors cite no prior work of their own), and no uniqueness theorem is imported from the authors' earlier papers. The evaluation compares against held-out GNM trajectories and external baselines (NWM, DINO-WM, OmniVLA, NoMaD) under identical CEM settings, so the 2.7x AOE reduction is an externally benchmarked comparison. The only caveat is that MCR's ranking signal and the GT-based displacement metrics measure the same construct, which complicates the 'without imitation' interpretation but does not make the prediction equivalent to its training inputs by construction. The stated limitations (static scenes, long-horizon error accumulation) are acknowledged rather than hidden.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

No new physical entities, particles, forces, or dimensions are introduced. The contributions are training objectives and an architecture configuration.

free parameters (8)
  • lambda_cos = 0.5
    Weight for cosine term in Eq. (3); set by hand, no sensitivity analysis.
  • MCR noise levels sigma = 0.05, 0.15, 0.35
    Perturbation strengths used to generate ranked candidates in Sec. 3.3.2; chosen by hand.
  • MCR AR(1) correlation rho = 0.7
    Correlation for temporally smooth perturbations; chosen by hand.
  • MCR margin constants = max(0.02, 0.5*abs(d_i - d_j))
    Gap-proportional margin in Eq. (6); chosen by hand.
  • Rollout counter-curriculum schedule = K 2 to 8, LR 1e-4 to 5e-6, batch 128 to 32
    Table 3; manually designed schedule to avoid gradient collapse.
  • MCR phase budget = 1000 steps, LR 1e-5, batch 8
    Appendix B.3; brief tuning phase, no sensitivity analysis.
  • CEM hyperparameters = 128 samples, 4 elites, 6 iterations, rho 0.5, K 6, S 4
    Appendix B.5; manually selected planning settings.
  • ACT weight lambda_act = 0.25
    Appendix B.4; used in investigated variant, not in final model.
assumptions (5)
  • domain assumption Frozen DINO-family features are a sufficient substrate for predicting future observations and for planning.
    Inherited from DINO-WM and 'Back to the features' [5][6]; used in Section 3.2 without re-validation in this paper.
  • domain assumption Cosine distance between pooled predicted embedding and goal embedding is a valid planning cost for image-goal navigation.
    Defined in Section 3.3.2 and Eq. (8); central to the method, not independently benchmarked against other cost functions.
  • domain assumption The ground-truth trajectory recorded in GNM is a good reference for ranking action sequences: larger dead-reckoned deviation from GT implies worse goal-reaching.
    Used to construct MCR pairs in Section 3.3.2; alternate paths to the goal may violate this ordering.
  • domain assumption CEM with 128 samples, 4 elites, and 6 iterations adequately solves the latent-space planning objective.
    Section 3.4; no sensitivity analysis or guarantee of global optimality.
  • standard math Standard ERM generalization: training on the GNM training portion transfers to the held-out 1,316-trajectory split.
    Section 4.1; standard assumption, but exact split is not specified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Latent World Models with Monotone Planning Costs for Image-Goal Navigation." pith.science (2026). https://pith.science/paper/VWQ25VM3

@misc{pith2026260809073,
  author       = {Pith},
  title        = {Pith review of: Latent World Models with Monotone Planning Costs for Image-Goal Navigation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VWQ25VM3}},
  note         = {Machine review of arXiv:2608.09073}
}
abstract

Image-goal navigation with latent world models requires not only accurate future prediction, but also a planning cost that reliably ranks candidate action sequences. We define the cost as the cosine distance between the predicted future embedding and the goal embedding, and show that poor cost ordering can mislead sampling-based planners such as Cross-Entropy Method (CEM). To address this, we propose a latent world model built on a frozen DINO-family encoder and train it with two complementary objectives. An autoregressive rollout loss reduces the gap between training and multi-step planning rollouts, while a Monotone Cost Ranking (MCR) loss directly encourages increasingly perturbed action sequences to receive higher planning costs. We also study InfoNCE-based action-contrastive training and find that temporal permutation negatives distort the latent geometry and degrade planning performance. On the GNM navigation dataset, our method outperforms Navigation World Models (NWM), DINO-WM, OmniVLA, and NoMaD, achieving state-of-the-art image-goal navigation performance while reducing orientation error by $2.7\times$ over the same-encoder DINO WM baseline. We also deploy the model zero-shot on a physical robot, where it follows goal-directed paths in unseen indoor and outdoor environments.

Figures

Figures reproduced from arXiv: 2608.09073 by the authors.

Figure 1
Figure 1. Latent world model with monotone planning cost. (Architecture, top.) A frozen DINO￾family encoder maps the current observation o1 to patch features z1; an AdaLN ViT predictor then rolls out H steps autoregressively, conditioned on actions (a1, . . . , aH), to produce embeddings zˆ2, . . . , zˆH entirely in latent space. The planning cost compares the predicted final embedding to the goal embedding zg (encoded from o… view at source ↗
Figure 2
Figure 2. Qualitative image-goal navigation example (held-out set). From the start (a) and goal (b) [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Qualitative comparison of zero-shot real-world navigation rollouts in outdoor and indoor [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Real-world robot platform and deployment pipeline. (a) Clearpath Husky A200 platform [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 12 canonical work pages

  1. [1]

    D. Shah, A. Sridhar, N. Dashora, K. Stachowicz, K. Black, N. Hirose, and S. Levine. ViNT: A foundation model for visual navigation. InConference on Robot Learning (CoRL), 2023. URL https://arxiv.org/abs/2306.14846

    arXiv 2023

  2. [2]

    Sridhar, D

    A. Sridhar, D. Shah, C. Glossop, and S. Levine. NoMaD: Goal masked diffusion policies for navigation and exploration. In2024 IEEE International Conference on Robotics and Automation (ICRA), 2024. URLhttps://arxiv.org/abs/2310.07896

    arXiv 2024

  3. [3]

    Hirose, C

    N. Hirose, C. Glossop, D. Shah, and S. Levine. OmniVLA: An omni-modal vision- language-action model for robot navigation.ArXiv, abs/2509.19480, 2025. URL https: //api.semanticscholar.org/CorpusID:281505165

    arXiv 2025

  4. [4]

    A. Bar, G. Zhou, D. Tran, T. Darrell, and Y . LeCun. Navigation world models.2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 15791–15801, 2024. URLhttps://api.semanticscholar.org/CorpusID:274465127

    work page 2025

  5. [5]

    G. Zhou, H. Pan, Y . LeCun, and L. Pinto. DINO-WM: World models on pre-trained visual features enable zero-shot planning. InInternational Conference on Machine Learning (ICML),

    work page

  6. [6]

    Baldassarre, M

    F. Baldassarre, M. Szafraniec, B. Terver, V . Khalidov, F. Massa, Y . LeCun, P. Labatut, M. Seitzer, and P. Bojanowski. Back to the features: DINO as a foundation for video world mod- els.ArXiv, abs/2507.19468, 2025. URL https://api.semanticscholar.org/CorpusID: 280045745

    arXiv 2025

  7. [7]

    D. Shah, A. Sridhar, A. Bhorkar, N. Hirose, and S. Levine. GNM: A general navigation model to drive any robot. In2023 IEEE International Conference on Robotics and Automation (ICRA),

    work page

  8. [8]

    W. Liu, H. Zhao, C. Li, J. Biswas, B. Okal, P. Goyal, Y . Chang, and S. Pouya. X-MOBILITY: End-to-end generalizable navigation via world modeling. In2025 IEEE International Con- ference on Robotics and Automation (ICRA), 2025. URL https://arxiv.org/abs/2410. 17491

    work page 2025

Show all 42 references
  1. [9]

    Balestriero and Y

    R. Balestriero and Y . LeCun. LeJEPA: Provable and scalable self-supervised learning without the heuristics.ArXiv, abs/2511.08544, 2025. URL https://api.semanticscholar.org/ CorpusID:282922448

    2025 arXiv

  2. [10]

    Bengio, O

    S. Bengio, O. Vinyals, N. Jaitly, and N. Shazeer. Scheduled sampling for sequence predic- tion with recurrent neural networks. InAdvances in Neural Information Processing Systems (NeurIPS), 2015. URLhttps://arxiv.org/abs/1506.03099

    2015 arXiv

  3. [11]

    Peebles and S

    W. Peebles and S. Xie. Scalable diffusion models with transformers. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 4195–4205, October 2023

    2023

  4. [12]

    Glossop, W

    C. Glossop, W. Chen, A. Bhorkar, D. Shah, and S. Levine. CAST: Counterfactual labels improve instruction following in vision-language-action models.ArXiv, abs/2508.13446, 2025. URL https://api.semanticscholar.org/CorpusID:280686229

    2025 arXiv

  5. [13]

    X. Liu, J. Li, Y . Jiang, N. Sujay, Z. Yang, J. Zhang, J. Abanes, J. Zhang, and C. Feng. Citywalker: Learning embodied urban navigation from web-scale videos.2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 6875–6885, 2024. URL https: //api.sem...

    2025

  6. [14]

    X. Chen, J. Guo, T. He, C. Zhang, P. Zhang, D. Yang, L. Zhao, and J. Bian. IGOR: Image-goal representations are the atomic control units for foundation models in embodied ai. 2024. URL https://api.semanticscholar.org/CorpusID:273811367

    2024

  7. [15]

    X. Luo, A. Wu, H. Han, X. Wan, W. Zhang, L. Shu, and R. Wang. Stepnav: Structured trajectory priors for efficient and multimodal visual navigation.ArXiv, abs/2602.02590, 2026. URL https://api.semanticscholar.org/CorpusID:285275552

    2026

  8. [16]

    S. Gode, A. Nayak, D. N. Oliveira, M. Krawez, C. Schmid, and W. Burgard. Flownav: Combining flow matching and depth priors for efficient navigation.2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 17762–17768, 2024. URL https: //api.semant...

    2025

  9. [17]

    Z. Chen, Y . Guo, Z. Chu, M. Luo, Y . Shen, M. Sun, J. Hu, S. Xie, K. Yang, P. Shi, Z. Gu, L. Liu, H. Han, X. Wu, M. Xu, and Y . Zhang. Socialnav: Training human-inspired foundation model for socially-aware embodied navigation.ArXiv, abs/2511.21135, 2025. URL https: //api.sema...

    2025

  10. [18]

    Suomela, N

    L. Suomela, N. Takahata, S. K. Arachchige, H. Edelman, and J.-K. K¨am¨ar¨ainen. Data scaling for navigation in unknown environments.IEEE Robotics and Automation Letters, 11:6114–6121,

  11. [19]

    H. He, Y . Ma, W. Wu, and B. Zhou. From seeing to experiencing: Scaling navigation foundation models with reinforcement learning.ArXiv, abs/2507.22028, 2025. URL https://api. semanticscholar.org/CorpusID:280337962

    2025 arXiv

  12. [20]

    Sobal, W

    V . Sobal, W. Zhang, K. Cho, R. Balestriero, T. G. J. Rudner, and Y . LeCun. Learn- ing from reward-free offline data: A case for planning with latent dynamics models. ArXiv, abs/2502.14819, 2025. URL https://api.semanticscholar.org/CorpusID: 276482758

    2025

  13. [21]

    Terver, T.-Y

    B. Terver, T.-Y . Yang, J. Ponce, A. Bardes, and Y . LeCun. What drives success in physical planning with joint-embedding predictive world models?ArXiv, abs/2512.24497, 2025. URL https://api.semanticscholar.org/CorpusID:284351435

    2025 arXiv

  14. [22]

    Assran, N

    M. Assran, N. Ballas, A. Bardes, Y . LeCun, M. Rabbat, et al. V-JEPA 2: Self-supervised video models enable understanding, prediction and planning, 2025. URL https://arxiv.org/ abs/2506.09985

    2025 arXiv

  15. [23]

    J. Sun, W. Zhang, Z. Qi, S. Ren, Z. Liu, H. Zhu, G. Sun, X. Jin, and Z. Chen. VLA-JEPA: Enhancing vision-language-action model with latent world model.ArXiv, abs/2602.10098,

  16. [24]

    Garrido, T

    Q. Garrido, T. Nagarajan, B. Terver, N. Ballas, Y . LeCun, and M. Rabbat. Learning la- tent action world models in the wild.ArXiv, abs/2601.05230, 2026. URL https://api. semanticscholar.org/CorpusID:284543722

    2026

  17. [25]

    L. Maes, Q. L. Lidec, D. Scieur, Y . LeCun, and R. Balestriero. LeWorldModel: Stable end-to-end joint-embedding predictive architecture from pixels. 2026. URL https://api. semanticscholar.org/CorpusID:286751061

    2026

  18. [26]

    Zhang, B

    W. Zhang, B. Terver, A. Zholus, S. Chitnis, H. Sutaria, M. Assran, R. Balestriero, A. Bar, A. Bardes, Y . LeCun, and N. Ballas. Hierarchical planning with latent world models, 2026. URLhttps://arxiv.org/abs/2604.03208

    2026 arXiv

  19. [27]

    URLhttps://api.semanticscholar.org/CorpusID:285462792

  20. [28]

    B. Hou, G. Li, J. Jia, et al. World model for robot learning: A comprehensive survey, 2026. URLhttps://arxiv.org/abs/2605.00080

    2026 arXiv

  21. [29]

    S. Ross, G. Gordon, and D. Bagnell. A reduction of imitation learning and structured prediction to no-regret online learning. InProceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics (AISTATS), 2011. URL https://arxiv.org/abs/1011. 0686

    2011

  22. [30]

    Hafner, T

    D. Hafner, T. Lillicrap, M. Norouzi, and J. Ba. Mastering diverse domains through world models.ArXiv, 2023. URLhttps://arxiv.org/abs/2301.04104

    2023 arXiv

  23. [31]

    Huang, J

    S. Huang, J. Wu, Q. Zhou, S. Miao, and M. Long. Vid2world: Crafting video diffusion models to interactive world models.ArXiv, abs/2505.14357, 2025. URL https://api. semanticscholar.org/CorpusID:278768612. 10

    2025

  24. [32]

    Sermanet, C

    P. Sermanet, C. Lynch, Y . Chebotar, J. Hsu, E. Jang, S. Schaal, and S. Levine. Time-contrastive networks: Self-supervised learning from video. In2018 IEEE International Conference on Robotics and Automation (ICRA), 2018. URLhttps://arxiv.org/abs/1704.06888

    2018 arXiv

  25. [33]

    T. Chen, S. Kornblith, M. Norouzi, and G. Hinton. A simple framework for contrastive learning of visual representations. InInternational Conference on Machine Learning (ICML), 2020. URLhttps://arxiv.org/abs/2002.05709

    2020 arXiv

  26. [34]

    Laskin, A

    M. Laskin, A. Srinivas, and P. Abbeel. Curl: Contrastive unsupervised representations for reinforcement learning. InInternational conference on machine learning, pages 5639–5650. PMLR, 2020. URLhttps://arxiv.org/abs/2004.04136

    2020 arXiv

  27. [35]

    van den Oord, Y

    A. van den Oord, Y . Li, and O. Vinyals. Representation learning with contrastive predictive coding.ArXiv, 2018. URLhttps://arxiv.org/abs/1807.03748

    2018 arXiv

  28. [36]

    Florence, C

    P. Florence, C. Lynch, A. Zeng, O. Ramirez, A. Wahid, L. Downs, A. Wong, J. Lee, I. Mordatch, and J. Tompson. Implicit behavioral cloning. InConference on Robot Learning (CoRL), 2021. URLhttps://arxiv.org/abs/2109.00137

    2021 arXiv

  29. [37]

    Y . J. Ma, S. Sodhani, D. Jayaraman, O. Bastani, V . Kumar, and A. Zhang. VIP: Towards universal visual reward and representation via value-implicit pre-training. InInternational Conference on Learning Representations (ICLR), 2023. URLhttps://arxiv.org/abs/2210.00030

    2023 arXiv

  30. [38]

    D. Wang, H. Ye, Z. Liang, Z. Sun, Z. Lu, Y . Zhang, Y . Zhao, Y . Gao, M. Seegert, F. Sch¨afer, H. Qin, W. Li, L. Palmieri, F. Jahncke, M. Piccinini, and J. Betz. Target-bench: Can world models achieve mapless path planning with semantic targets?ArXiv, abs/2511.17792, 2025. UR...

    2025 arXiv

  31. [39]

    Eysenbach, T

    B. Eysenbach, T. Zhang, R. Salakhutdinov, and S. Levine. Contrastive learning as goal- conditioned reinforcement learning. InAdvances in Neural Information Processing Systems (NeurIPS), 2022. URLhttps://arxiv.org/abs/2206.07568

    2022 arXiv

  32. [2023]

    URLhttps://arxiv.org/abs/2210.03370

    arXiv

  33. [2025]

    URLhttps://arxiv.org/abs/2411.04983

    arXiv

  34. [2026]

    URLhttps://api.semanticscholar.org/CorpusID:284717654

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.