Pith. sign in

REVIEW 5 major objections 7 minor 50 references

Efficient Robotic Policy Learning via Latent Space Backward Planning

T0 review · 5 major / 7 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read Backward latent planning hits 88.6% on long-horizon robot tasks

desk verdict LBP's backward latent subgoal planning is a genuinely new and well-tested idea, but the evidence for its backward-over-forward advantage is confounded and needs a matched baseline before the strong claim can stand. read the letter →

arxiv 2505.06861 v2 pith:VT6CV6DQ submitted 2025-05-11 cs.RO cs.AIcs.CV

classification cs.ROcs.AIcs.CV
keywords latentspaceplanningbackwardsubgoalgenerationlong-horizonrobotmanipulationgoal-conditionedpolicylanguage-conditionedimitationlearningdiffusion
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 proposes Latent Backward Planning (LBP), a way to guide a robot policy through long, multi-stage tasks by predicting where the task ends and then working backward. Instead of generating future video frames or predicting subgoals forward step by step, LBP first maps the current image and language instruction to a latent final goal, then recursively produces intermediate subgoals closer to the current state. The paper argues that this reverses the usual direction of error accumulation: because every subgoal is anchored to the final goal, plans stay aligned with task completion. On the LIBERO-LONG benchmark LBP reports 88.6% average success with a three-step plan, above Seer (78.6%) and SuSIE (76.3%), and it outperforms baselines in the later stages of real-robot cup-manipulation tasks. The proposal matters because it offers a lightweight planner in latent space rather than heavy generative models, as a route to real-time long-horizon control.

What carries the argument

The central mechanism is the recursive backward subgoal chain. A goal predictor grounds the language instruction into a final latent goal $z_g$; a single subgoal predictor $f_w$, conditioned on current state $z_t$, language, and the previous subgoal $w_{i-1}$ (with $w_0 = z_g$), generates each nearer subgoal. The planning coefficient $\lambda = \frac{\Gamma(w_i) - t}{\Gamma(w_{i-1}) - t}$ controls how far each subgoal lies between the current state and the goal. A Perceiver-style cross-attention goal-fusion module compresses the resulting context sequence into one learnable token that conditions the diffusion policy. Predicting from the goal backward is what carries the claim that subgoals remain on-task and that error accumulation is controlled.

What would settle it

Run LBP on test episodes with object positions, target objects, or backgrounds shifted beyond the training distribution while keeping the instructions the same, and compare the predicted final latent goal against the true final state recorded by hindsight labeling and against downstream task success. If success drops sharply exactly on episodes where the final-goal prediction is wrong while the backward subgoal chain stays internally consistent with that wrong goal, the claim that backward planning mitigates compounding errors would be undermined because the error would simply relocate to the goal predictor.

Watch

Extended reading notes

Core claim

LBP's central claim is that backward subgoal prediction in latent space resolves the trilemma of planning efficiency, long-horizon consistency, and prediction accuracy that forward planners face. The model learns a goal predictor that, from the current latent state and language instruction, predicts the final latent goal; a unified subgoal predictor then recursively predicts intermediate subgoals, each conditioned on the current state, the language, and the previous subgoal that lies closer to the goal. Because the recursion works backward from the final goal toward the present, the number of prediction steps stays small and each predicted subgoal is supervised against ground-truth trajectory states, including a consistency term in which the predictor consumes its own previous predictions. The paper's evidence includes the LIBERO-LONG success rates and a subgoal prediction error comparison in which forward planning errors grow rapidly across task progress while backward errors stay low. The conclusion the authors draw is that grounding plans in a final goal fundamentally mitigates the compounding prediction errors of forward planning.

Load-bearing premise

The load-bearing assumption is that the model that predicts the final completed state from the current view and the instruction keeps working when the scene changes at test time; if that final-state prediction is wrong, every subgoal planned backward from it inherits the error.

Editorial extensions

If this is right

  • Planning in latent space with few subgoals (a final goal plus two intermediates) reaches 88.6% average success on LIBERO-LONG, so a lightweight MLP planner can replace heavier video- or image-editing planners without sacrificing accuracy.
  • Because backward planning keeps subgoal prediction error low while forward autoregressive error grows across task progress, long-horizon tasks should degrade less in their later stages.
  • The goal-fusion module's adaptive compression of subgoal contexts outperforms average pooling by 9.6% on LIBERO-LONG, indicating that policies need to weight near and far subgoals differently.
  • The recursive planning coefficient $\lambda$ shows little sensitivity between 0.5 and 0.75, suggesting backward planning reduces the burden of tuning subgoal spacing.
  • On real-robot tasks the largest margins appear at the final stages, as in Shift Cups where LBP scores 26.6 at the final stage while LCBC, GLCBC, and SuSIE all score 0.0.

Reading between the lines

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

  • Going beyond the paper, the backward-planning principle could transfer to other compact state representations such as keypoints or object-centric states, where the final-goal predictor would have an even smaller error surface than full latent images.
  • An extension not explored in the paper is backward replanning under mid-task disturbances: because replanning regenerates the goal and subgoals from the current state, LBP could be tested for closed-loop recovery after a perturbation such as an object being knocked out of place.
  • The paper leaves implicit that backward planning is only as good as the final-goal prediction; a natural test is whether injecting a deliberately wrong final goal causes the entire backward chain to stay consistently wrong, which would show where the residual error actually lives.
  • Another implication is that the method's efficiency depends on a frozen pretrained latent encoder, so the quality of planning may be bounded by the encoder's ability to represent task-relevant scene changes; improving the encoder could yield further gains without changing the planning recursion.
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

5 major / 7 minor

Summary. The paper proposes Latent Space Backward Planning (LBP), a method for language-conditioned robotic manipulation that plans in a frozen latent space by first predicting a final goal latent from the current observation and language instruction, and then recursively predicting intermediate subgoals backward from the goal toward the current state. A subgoal-conditioned diffusion policy with a Perceiver-style goal-fusion attention module is trained to use the planned subgoal sequence. The method is evaluated on the LIBERO-LONG simulation benchmark and on four real-robot long-horizon tasks, reporting higher average success rates and stage scores than baselines including Seer, SuSIE, LCBC, and GLCBC. The paper also includes ablations on the recursive planning coefficient λ, the number of subgoals, and the goal-fusion strategy, plus additional comparisons to forward and parallel planning paradigms in the appendices.

Significance. If the empirical results are reliable, LBP is a practically valuable contribution: it replaces expensive generative video or image planners with lightweight MLP predictors in latent space, potentially enabling real-time planning for long-horizon tasks. The backward coarse-to-fine planning idea is intuitive and well motivated, and the goal-fusion mechanism is shown to be useful through ablation. The paper is also unusually transparent in reporting implementation details and dataset augmentations, and it provides numerical tables for the real-robot experiments. However, the strongest conceptual claim, that backward planning 'fundamentally mitigates' compounding prediction errors, currently rests on a confounded comparison that does not isolate the direction of planning from goal conditioning or the number of recursive steps. The evaluation also lacks variance reporting, which weakens the significance of the headline margins. The core contribution is promising, but the evidence needs to be strengthened before the central mechanistic claim can be accepted.

major comments (5)
  1. [Section 5.2 and Figure 5] The comparison between the backward and forward planning paradigms is confounded. The forward planner is described as 'predict[ing] the subgoal 10 steps ahead at each iteration, autoregressively generating the entire subgoal sequence' and is not stated to be conditioned on the final latent goal zg, whereas LBP always conditions every subgoal on zg via Eqs. (3)-(5). The two paradigms also use different numbers of recursive steps (LBP: zg plus two subgoals; forward: ten autoregressive steps). The lower MSE of LBP in Figure 5 could therefore be entirely due to goal grounding or to the smaller number of recursive steps, rather than to the backward ordering. To support the Section 6 claim that backward planning 'fundamentally mitigates the compounding prediction errors,' the authors should compare forward and backward planners under matched conditions: identical goal conditioning, identical numbers of planning steps, and ideally a closed-loop success-rate comparison. As it stands, the central mechanistic claim is under-supported.
  2. [Section 5.1 vs. Appendix C] The paper gives inconsistent numbers of real-robot demonstrations. Section 5.1 states that 'All models are trained using 200 expert demonstrations for the task Move cups and Shift cups, and a total of 200 expert demonstrations for Stack 3 cups and Stack 4 cups,' while Appendix C states 'We collect 200 expert demonstrations each for tasks Move cups, Stack 3 cups, Stack 4 cups and Shift cups.' If Stack 3 cups and Stack 4 cups indeed share 200 demonstrations in total, the baselines on those tasks may be at a data disadvantage, undermining the comparison in Figure 4 and Tables 7-10. The authors must correct this inconsistency and clarify the actual per-task demonstration counts.
  3. [Table 1 and Figure 4] No measures of variance or seed multiplicity are reported anywhere in the main results. Table 1 reports only the average success rate of the top-3 checkpoints over 10 rollouts per task, with no standard deviations, confidence intervals, or per-seed values. Figure 4 and Tables 7-11 likewise report average stage scores without error bars. Given that the paper claims 'SOTA performance' and 'robustness,' the absence of any variance reporting makes it impossible to assess whether the reported margins over baselines (e.g., 88.6% vs. 78.6% for Seer in Table 1) are statistically meaningful. The authors should provide standard deviations across rollouts and ideally across multiple training seeds.
  4. [Section 4.2, Eq. (5), and Section 6] The claim that the recursive self-conditioning in Eq. (5) 'will suffer much less compounding error' is asserted without formal or empirical support that isolates the mechanism. The second term of Eq. (5) trains fw on its own previous predictions, which is a plausible scheduled-sampling heuristic, but the text provides no analysis of how prediction errors propagate through the λ-recursion, and the only direct evidence (Figure 5) is confounded as noted above. The Section 6 statement that LBP 'fundamentally mitigates the compounding prediction errors inherent in traditional forward planning approaches' should be tempered to a more defensible claim (e.g., 'reduces' or 'mitigates in the evaluated settings') unless the authors provide a direct error-propagation analysis or an additional matched closed-loop comparison.
  5. [Section 5.3 and Table 2] The hyperparameters λ and the number of planning steps are selected via ablations on LIBERO-LONG, which is the same benchmark used for the headline results in Table 1. For example, the configuration λ=0.5 with three subgoals is chosen because it gives 88.6% average success, while other configurations give lower values. This introduces a mild selection-on-test-set bias, as the reported SOTA number is the maximum over the hyperparameter grid. The authors should clarify whether the ablation was performed on a separate validation split, and should report the performance of the chosen configuration under multiple seeds to quantify the sensitivity of the headline result to hyperparameter selection.
minor comments (7)
  1. [Appendix A] The sentence 'The detailed architecture of our model is present in Figure 2' should read 'presented in Figure 2.'
  2. [Section 5.2 and Figure 5] For the forward-paradigm comparison, please specify exactly how many subgoals the forward planner predicts in total and how the 10-step-ahead prediction is iterated; also report the MSE curves separately per task, since aggregating across tasks may hide differences in the difficulty of the planning horizon.
  3. [Figure 5 and Figure 6] The MSE is computed over 3,000 sampled states, but no variance or confidence intervals are shown for the MSE curves; adding error bars would make the comparison more informative.
  4. [Appendix D] The 'parallel planning' comparison in Figure 6 suffers from the same confound as the forward comparison: the parallel planner is not described as being conditioned on the final latent goal zg, so it does not isolate the effect of planning direction.
  5. [Table 11] The generalization experiment on Shift cups only evaluates LBP under distracting objects and different backgrounds; LCBC is only reported in the base setting. For a fair robustness comparison, the same baselines should be evaluated under the altered conditions.
  6. [Appendix C] There is a typo in 'View augmentation always exsits' — 'exsits' should be 'exists.'
  7. [Section 5.1] The evaluation protocol of averaging over the 'top-3 checkpoints' should be justified; if the top checkpoints are selected using validation performance, this should be stated explicitly, and the sensitivity of the final result to this selection procedure should be discussed.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: LBP's goal and subgoal predictors are supervised by ground-truth latent states from expert trajectories and evaluated on external rollouts; the central derivation is not equivalent to its inputs.

full rationale

The derivation chain is self-contained. The final goal predictor fg is trained by maximizing log fg(zg|zt, φl) against hindsight latent goal labels from expert trajectories (Eq. 2), and the unified subgoal predictor fw is trained against ground-truth subgoals z_{λi} ⊂ τ plus its own previous predictions (Eq. 5), so neither objective defines its target in terms of the downstream policy's success. The policy (Eq. 6) consumes predicted latent subgoals and is evaluated by closed-loop success rates on LIBERO-LONG and real-robot tasks, which are external quantities not constructed from the training labels. The reported ablations (Tables 2–3) are honest: removing the planner or the goal-fusion module changes architecture and performance, and λ/subgoal-count selection on the same benchmark is ordinary hyperparameter tuning that mildly inflates optimism but does not make any reported number a fitted rename of an input. Self-citations to DecisionNCE (Li et al., 2024) and to the coarse-to-fine decision-making paper (Wang et al., 2025) are not load-bearing: DecisionNCE is a frozen external encoder and the paper also reports a SigLIP variant (85.0%), and the coarse-to-fine reference is only inspirational framing. The forward-vs-backward MSE comparison in Section 5.2/Figure 5 is confounded (the forward baseline is not conditioned on the final latent goal and uses 10-step autoregressive subgoals), but that is a validity threat to the mechanistic claim, not a circularity in which an output equals an input by construction.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the frozen latent space being a valid planning space, on the goal predictor staying accurate at test time, and on recursive subgoal prediction remaining stable. These are domain assumptions rather than independently evidenced facts; lambda and the number of subgoals are tuned on the evaluation benchmark.

free parameters (2)
  • recursive planning coefficient lambda = 0.5
    Chosen by hand and ablated in Table 2; controls how close each predicted subgoal is to the final goal as a fraction of the remaining horizon.
  • number of planning steps (final goal plus subgoals) = 3 (zg, w1, w2)
    Selected by ablation on LIBERO-LONG (Table 2): zg plus two subgoals gave 88.6, while fewer or more subgoals reduced performance.
assumptions (3)
  • domain assumption Frozen latent encoders (DecisionNCE/SigLIP) provide a planning space where latent states at time-interpolation points are meaningful subgoals that guide the low-level policy.
    Eq. (5) defines subgoal targets z_lambda_i := z_ceil((1-lambda_i)t + lambda_i H) from demonstration trajectories and relies on these being useful training signals.
  • domain assumption Hindsight labeling from expert demonstrations is a correct grounding of task completion: the final latent state is the goal implied by the language instruction and current state.
    Eq. (2) trains fg to predict the final latent state from (zt, phi_l), assuming this mapping is well-defined for each task.
  • ad hoc to paper Recursive self-conditioning at test time remains stable because the second term of Eq. (5) trains fw on its own previous predictions.
    Section 4.2 claims the recursion will suffer much less compounding error but provides no formal bound and does not quantify fg error propagation into the chain.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Efficient Robotic Policy Learning via Latent Space Backward Planning." pith.science (2026). https://pith.science/paper/VT6CV6DQ

@misc{pith2026250506861,
  author       = {Pith},
  title        = {Pith review of: Efficient Robotic Policy Learning via Latent Space Backward Planning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VT6CV6DQ}},
  note         = {Machine review of arXiv:2505.06861}
}
read the original abstract

Current robotic planning methods often rely on predicting multi-frame images with full pixel details. While this fine-grained approach can serve as a generic world model, it introduces two significant challenges for downstream policy learning: substantial computational costs that hinder real-time deployment, and accumulated inaccuracies that can mislead action extraction. Planning with coarse-grained subgoals partially alleviates efficiency issues. However, their forward planning schemes can still result in off-task predictions due to accumulation errors, leading to misalignment with long-term goals. This raises a critical question: Can robotic planning be both efficient and accurate enough for real-time control in long-horizon, multi-stage tasks? To address this, we propose a Latent Space Backward Planning scheme (LBP), which begins by grounding the task into final latent goals, followed by recursively predicting intermediate subgoals closer to the current state. The grounded final goal enables backward subgoal planning to always remain aware of task completion, facilitating on-task prediction along the entire planning horizon. The subgoal-conditioned policy incorporates a learnable token to summarize the subgoal sequences and determines how each subgoal guides action extraction. Through extensive simulation and real-robot long-horizon experiments, we show that LBP outperforms existing fine-grained and forward planning methods, achieving SOTA performance. Project Page: https://lbp-authors.github.io

Figures

Figures reproduced from arXiv: 2505.06861 by the authors.

Figure 1
Figure 1. Illustration of latent space backward planning. 2. Related Works Video Planning. A significant body of research has ex￾plored video generation as planners for visuomotor con￾trol (Pertsch et al., 2020; Du et al., 2023; Ajay et al., 2024; Hu et al., 2024; Wu et al., 2024; Bharadhwaj et al., 2024). Approaches such as UniPi (Du et al., 2023) and HiP (Ajay et al., 2024) generate actions using inverse dynamics models fro… view at source ↗
Figure 2
Figure 2. Overall framework architecture of LBP. reflect task progression, while long subgoal sequences are prone to compounding prediction errors that lead to off-task behaviors deviating from the intended task goals. To address this, we begin by predicting the first subgoal w1 from the current state zt, final goal zg, and language instruction ϕl in latent space, with the optimization objective: max f 1 w X τ∈Dz X 1≤t≤H Ep(w… view at source ↗
Figure 3
Figure 3. Left: the entire desktop environment setups of real-world experiments contains a 6 DoF AIRBOT arm and three Logitech C922PRO cameras with different views; Right: (1) Move cups: move both brown cups in front of the white ones; (2) Stack cups: stack all paper cups together; (3) Shift cups: shift all the paper cups to another plate, in a clockwise direction. Stage I Stage II Stage III Stage IV Stage V Shift cups LCBC S… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Real-world main results. We evaluate LCBC, GLCBC, SuSIE and LBP in aforementioned 4 tasks. The metric ”Avg. Score” measures the average score for each stage. We observe that while LBP slightly outperforms other strong baselines at the early stages, LBP wins by a fairly…
Figure 5
Figure 5. Figure 5: Mean Squared Errors (MSE) between predicted subgoals and corresponding ground truths under forward and backward paradigm [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Mean Squared Errors (MSE) between predicted subgoals and corresponding ground truths in parallel, forward and backward planning [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 37 canonical work pages

  1. [1]

    Is conditional generative modeling all you need for decision-making? In International Conference on Learning Representations, 2023

    Ajay, A., Du, Y., Gupta, A., Tenenbaum, J., Jaakkola, T., and Agrawal, P. Is conditional generative modeling all you need for decision-making? In International Conference on Learning Representations, 2023

  2. [2]

    Compositional foundation models for hierarchical planning

    Ajay, A., Han, S., Du, Y., Li, S., Gupta, A., Jaakkola, T., Tenenbaum, J., Kaelbling, L., Srivastava, A., and Agrawal, P. Compositional foundation models for hierarchical planning. In Advances in Neural Information Processing Systems, 2024

  3. [3]

    Gen2act: Human video generation in novel scenarios enables generalizable robot manipulation

    Bharadhwaj, H., Dwibedi, D., Gupta, A., Tulsiani, S., Doersch, C., Xiao, T., Shah, D., Xia, F., Sadigh, D., and Kirmani, S. Gen2act: Human video generation in novel scenarios enables generalizable robot manipulation. arXiv preprint arXiv:2409.16283, 2024

  4. [4]

    R., Finn, C., Kumar, A., and Levine, S

    Black, K., Nakamoto, M., Atreya, P., Walke, H. R., Finn, C., Kumar, A., and Levine, S. Zero-shot robotic manipulation with pre-trained image-editing diffusion models. In International Conference on Learning Representations, 2024

  5. [5]

    Goal-conditioned reinforcement learning with imagined subgoals

    Chane-Sane, E., Schmid, C., and Laptev, I. Goal-conditioned reinforcement learning with imagined subgoals. In International Conference on Learning Representations, 2021

  6. [6]

    Diffusion policy: Visuomotor policy learning via action diffusion

    Chi, C., Feng, S., Du, Y., Xu, Z., Cousineau, E., Burchfiel, B., and Song, S. Diffusion policy: Visuomotor policy learning via action diffusion. In Robotics: Science and Systems, 2023

  7. [7]

    Learning universal policies via text-guided video generation

    Du, Y., Yang, S., Dai, B., Dai, H., Nachum, O., Tenenbaum, J., Schuurmans, D., and Abbeel, P. Learning universal policies via text-guided video generation. In Advances in Neural Information Processing Systems, 2023

  8. [8]

    Rvs: What is essential for offline RL via supervised learning? In International Conference on Learning Representations, 2022

    Emmons, S., Eysenbach, B., Kostrikov, I., and Levine, S. Rvs: What is essential for offline RL via supervised learning? In International Conference on Learning Representations, 2022

Show all 50 references
  1. [9]

    R., and Levine, S

    Eysenbach, B., Salakhutdinov, R. R., and Levine, S. Search on the replay buffer: Bridging planning and reinforcement learning. In Advances in Neural Information Processing Systems, 2019

  2. [10]

    Planning to practice: Efficient online fine-tuning by composing goals in latent space

    Fang, K., Yin, P., Nair, A., and Levine, S. Planning to practice: Efficient online fine-tuning by composing goals in latent space. In IEEE/RSJ International Conference on Intelligent Robots and Systems, 2022

  3. [11]

    M., Eysenbach, B., and Levine, S

    Ghosh, D., Gupta, A., Reddy, A., Fu, J., Devin, C. M., Eysenbach, B., and Levine, S. Learning to reach goals via iterated supervised learning. In International Conference on Machine Learning, 2021

  4. [12]

    G., and Levine, S

    Hansen-Estruch, P., Kostrikov, I., Janner, M., Kuba, J. G., and Levine, S. Idql: Implicit q-learning as an actor-critic method with diffusion policies. arXiv preprint arXiv:2304.10573, 2023

  5. [13]

    Deep residual learning for image recognition

    He, K., Zhang, X., Ren, S., and Sun, J. Deep residual learning for image recognition. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2016

  6. [14]

    Video prediction policy: A generalist robot policy with predictive visual representations

    Hu, Y., Guo, Y., Wang, P., Chen, X., Wang, Y.-J., Zhang, J., Sreenath, K., Lu, C., and Chen, J. Video prediction policy: A generalist robot policy with predictive visual representations. arXiv preprint arXiv:2412.14803, 2024

  7. [15]

    Subgoal diffuser: Coarse-to-fine subgoal generation to guide model predictive control for robot manipulation

    Huang, Z., Lin, Y., Yang, F., and Berenson, D. Subgoal diffuser: Coarse-to-fine subgoal generation to guide model predictive control for robot manipulation. In IEEE International Conference on Robotics and Automation, 2024

  8. [16]

    Perceiver: General perception with iterative attention

    Jaegle, A., Gimeno, F., Brock, A., Vinyals, O., Zisserman, A., and Carreira, J. Perceiver: General perception with iterative attention. In International Conference on Machine Learning, 2021

  9. [17]

    Planning with diffusion for flexible behavior synthesis

    Janner, M., Du, Y., Tenenbaum, J., and Levine, S. Planning with diffusion for flexible behavior synthesis. In International Conference on Machine Learning, 2022

  10. [18]

    and Kuo, Y.-L

    Kang, X. and Kuo, Y.-L. Incorporating task progress knowledge for subgoal generation in robotic manipulation through image edits. In IEEE/CVF Winter Conference on Applications of Computer Vision, 2025

  11. [19]

    J., Pertsch, K., Karamcheti, S., Xiao, T., Balakrishna, A., Nair, S., Rafailov, R., Foster, E

    Kim, M. J., Pertsch, K., Karamcheti, S., Xiao, T., Balakrishna, A., Nair, S., Rafailov, R., Foster, E. P., Sanketi, P. R., Vuong, Q., Kollar, T., Burchfiel, B., Tedrake, R., Sadigh, D., Levine, S., Liang, P., and Finn, C. Open VLA : An open-source vision-language-action model....

  12. [20]

    Hierarchical reinforcement learning with hindsight

    Levy, A., Platt, R., and Saenko, K. Hierarchical reinforcement learning with hindsight. In International Conference on Learning Representations, 2019

  13. [21]

    Decision NCE : Embodied multimodal representations via implicit preference learning

    Li, J., Zheng, J., Zheng, Y., Mao, L., Hu, X., Cheng, S., Niu, H., Liu, J., Liu, Y., Liu, J., et al. Decision NCE : Embodied multimodal representations via implicit preference learning. In International Conference on Machine Learning, 2024

  14. [22]

    Robo-mutual: Robotic multimodal task specification via unimodal learning

    Li, J., Wang, Z., Zheng, J., Zhou, X., Wang, G., Song, G., Liu, Y., Liu, J., Zhang, Y.-Q., Yu, J., and Zhan, X. Robo-mutual: Robotic multimodal task specification via unimodal learning. In IEEE International Conference on Robotics and Automation, 2025

  15. [23]

    Libero: Benchmarking knowledge transfer for lifelong robot learning

    Liu, B., Zhu, Y., Gao, C., Feng, Y., Liu, Q., Zhu, Y., and Stone, P. Libero: Benchmarking knowledge transfer for lifelong robot learning. In Advances in Neural Information Processing Systems, 2024

  16. [24]

    and Sermanet, P

    Lynch, C. and Sermanet, P. Language conditioned imitation learning over unstructured data. In Robotics: Science and Systems, 2021

  17. [25]

    Why does hierarchy (sometimes) work so well in reinforcement learning? arXiv preprint arXiv:1909.10618, 2019

    Nachum, O., Tang, H., Lu, X., Gu, S., Lee, H., and Levine, S. Why does hierarchy (sometimes) work so well in reinforcement learning? arXiv preprint arXiv:1909.10618, 2019

  18. [26]

    and Finn, C

    Nair, S. and Finn, C. Hierarchical foresight: Self-supervised learning of long-horizon tasks via visual subgoal generation. In International Conference on Learning Representations, 2020

  19. [27]

    Goal-aware prediction: Learning to model what matters

    Nair, S., Savarese, S., and Finn, C. Goal-aware prediction: Learning to model what matters. In International Conference on Machine Learning, 2020

  20. [28]

    Planning with goal-conditioned policies

    Nasiriany, S., Pong, V., Lin, S., and Levine, S. Planning with goal-conditioned policies. In Advances in Neural Information Processing Systems, 2019

  21. [29]

    Hiql: Offline goal-conditioned rl with latent states as actions

    Park, S., Ghosh, D., Eysenbach, B., and Levine, S. Hiql: Offline goal-conditioned rl with latent states as actions. In Advances in Neural Information Processing Systems, 2024

  22. [30]

    Film: Visual reasoning with a general conditioning layer

    Perez, E., Strub, F., De Vries, H., Dumoulin, V., and Courville, A. Film: Visual reasoning with a general conditioning layer. In AAAI Conference on Artificial Intelligence, 2018 a

  23. [31]

    Film: Visual reasoning with a general conditioning layer

    Perez, E., Strub, F., De Vries, H., Dumoulin, V., and Courville, A. Film: Visual reasoning with a general conditioning layer. In AAAI Conference on Artificial Intelligence, 2018 b

  24. [32]

    Long-horizon visual planning with goal-conditioned hierarchical predictors

    Pertsch, K., Rybkin, O., Ebert, F., Zhou, S., Jayaraman, D., Finn, C., and Levine, S. Long-horizon visual planning with goal-conditioned hierarchical predictors. In Advances in Neural Information Processing Systems, 2020

  25. [33]

    W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al

    Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al. Learning transferable visual models from natural language supervision. In International Conference on Learning Representations, 2021

  26. [34]

    Mutex: Learning unified policies from multimodal task specifications

    Shah, R., Mart \' n-Mart \' n, R., and Zhu, Y. Mutex: Learning unified policies from multimodal task specifications. In Conference on Robot Learning, 2023

  27. [35]

    Visual autoregressive modeling: Scalable image generation via next-scale prediction

    Tian, K., Jiang, Y., Yuan, Z., PENG, B., and Wang, L. Visual autoregressive modeling: Scalable image generation via next-scale prediction. In Advances in Neural Information Processing Systems, 2024

  28. [36]

    Predictive inverse dynamics models are scalable learners for robotic manipulation

    Tian, Y., Yang, S., Zeng, J., Wang, P., Lin, D., Dong, H., and Pang, J. Predictive inverse dynamics models are scalable learners for robotic manipulation. In International Conference on Learning Representations, 2025

  29. [37]

    D., Chang, M., Janner, M., Finn, C., Wu, J., Tenenbaum, J., and Levine, S

    Veerapaneni, R., Co-Reyes, J. D., Chang, M., Janner, M., Finn, C., Wu, J., Tenenbaum, J., and Levine, S. Entity abstraction in visual model-based reinforcement learning. In Conference on Robot Learning, 2020

  30. [38]

    Mimicplay: Long-horizon imitation learning by watching human play

    Wang, C., Fan, L., Sun, J., Zhang, R., Fei-Fei, L., Xu, D., Zhu, Y., and Anandkumar, A. Mimicplay: Long-horizon imitation learning by watching human play. In Conference on Robot Learning, 2023

  31. [39]

    Are expressive models truly necessary for offline RL ? In AAAI Conference on Artificial Intelligence, 2025

    Wang, G., Niu, H., Li, J., Jiang, L., HU, J., and Zhan, X. Are expressive models truly necessary for offline RL ? In AAAI Conference on Artificial Intelligence, 2025

  32. [40]

    V., Zhou, D., et al

    Wei, J., Wang, X., Schuurmans, D., Bosma, M., Xia, F., Chi, E., Le, Q. V., Zhou, D., et al. Chain-of-thought prompting elicits reasoning in large language models. In Advances in Neural Information Processing Systems, 2022

  33. [41]

    Any-point trajectory modeling for policy learning

    Wen, C., Lin, X., So, J., Chen, K., Dou, Q., Gao, Y., and Abbeel, P. Any-point trajectory modeling for policy learning. In Robotics: Science and Systems, 2024

  34. [42]

    Unleashing large-scale video generative pre-training for visual robot manipulation

    Wu, H., Jing, Y., Cheang, C., Chen, G., Xu, J., Li, X., Liu, M., Li, H., and Kong, T. Unleashing large-scale video generative pre-training for visual robot manipulation. In International Conference on Learning Representations, 2024

  35. [43]

    Masked visual pre-training for motor control

    Xiao, T., Radosavovic, I., Darrell, T., and Malik, J. Masked visual pre-training for motor control. arXiv preprint arXiv:2203.06173, 2022

  36. [44]

    A policy-guided imitation approach for offline reinforcement learning

    Xu, H., Li, J., Li, J., and Zhan, X. A policy-guided imitation approach for offline reinforcement learning. In Advances in Neural Information Processing Systems, 2022

  37. [45]

    Learning manipulation by predicting interaction

    Zeng, J., Bu, Q., Wang, B., Xia, W., Chen, L., Dong, H., Song, H., Wang, D., Hu, D., Luo, P., et al. Learning manipulation by predicting interaction. In Robotics: Science and Systems, 2024

  38. [46]

    Sigmoid loss for language image pre-training

    Zhai, X., Mustafa, B., Kolesnikov, A., and Beyer, L. Sigmoid loss for language image pre-training. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023

  39. [47]

    Z., Kumar, V., Levine, S., and Finn, C

    Zhao, T. Z., Kumar, V., Levine, S., and Finn, C. Learning fine-grained bimanual manipulation with low-cost hardware. arXiv preprint arXiv:2304.13705, 2023

  40. [48]

    Instruction-guided visual masking

    Zheng, J., Li, J., Cheng, S., Zheng, Y., Li, J., Liu, J., Liu, Y., Liu, J., and Zhan, X. Instruction-guided visual masking. In Advances in Neural Information Processing Systems, 2024

  41. [49]

    Universal actions for enhanced embodied foundation models

    Zheng, J., Li, J., Liu, D., Zheng, Y., Wang, Z., Ou, Z., Liu, Y., Liu, J., Zhang, Y.-Q., and Zhan, X. Universal actions for enhanced embodied foundation models. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025

  42. [50]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

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