pith. machine review for the scientific record. sign in

arxiv: 2605.11479 · v1 · submitted 2026-05-12 · 💻 cs.RO · cs.AI

Recognition: 2 theorem links

· Lean Theorem

Offline Policy Evaluation for Manipulation Policies via Discounted Liveness Formulation

Colton Crosby, Hao Wang, Joshua Bowden, Somil Bansal

Pith reviewed 2026-05-13 02:10 UTC · model grok-4.3

classification 💻 cs.RO cs.AI
keywords offline policy evaluationmanipulation policiesliveness formulationBellman operatortruncation biassparse rewardsrobotic task completion
0
0 comments X

The pith

A discounted liveness Bellman operator produces a conservative fixed-point value function for offline evaluation of manipulation policies that remains accurate under finite rollouts.

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

The paper reframes offline policy evaluation as a task-completion problem to address sparse rewards, non-monotonic recovery behaviors, and truncation bias from short rollouts in robotic manipulation. Standard Bellman updates break down under these conditions because they rely on infinite-horizon assumptions that do not hold in practice. The proposed liveness-based operator directly encodes task progression into the update rule, yielding a conservative value estimate that converges even when episodes are cut short. Theoretical analysis establishes contraction guarantees for the operator, ensuring the fixed point exists and can be computed reliably. Experiments on simulated tasks with vision-language-action and diffusion policies, plus a cloth-folding demonstration set, show improved alignment with actual task success compared to TD(0) and Monte Carlo baselines.

Core claim

The paper claims that incorporating a discounted liveness property into the Bellman operator allows policy evaluation to treat the problem as determining whether a task will eventually complete. This produces a conservative value function whose fixed point remains stable under finite-horizon truncation and captures non-monotonic progress without requiring auxiliary labels or monotonicity assumptions.

What carries the argument

The liveness-based Bellman operator that augments the standard backup with a term encoding task-completion progress.

If this is right

  • Value estimates remain conservative and monotonically related to true task success even when rollouts are truncated.
  • The operator converges under the same conditions as standard discounted Bellman updates because it is a contraction.
  • Evaluation accuracy improves for policies that exhibit recovery behaviors, as shown on vision-language-action models and diffusion policies.
  • The same formulation applies to human demonstration data without modification, as demonstrated on cloth folding.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The approach could be combined with existing offline RL datasets to rank policies before real-world deployment without lengthening simulations.
  • It suggests a route to conservative safety checks in manipulation where overestimating progress is costly.
  • Extensions to multi-task settings would require only redefining the liveness predicate for each task.

Load-bearing premise

Task completion can be expressed as a liveness property that fits directly inside the Bellman operator without extra progress labels or assumptions that fail during recovery.

What would settle it

Compare the learned values against ground-truth completion rates on rollouts long enough to reach success or clear failure; the estimates should stay below true rates yet increase toward them as horizon length grows, unlike standard methods that oscillate or drop sharply near truncation.

Figures

Figures reproduced from arXiv: 2605.11479 by Colton Crosby, Hao Wang, Joshua Bowden, Somil Bansal.

Figure 1
Figure 1. Figure 1: In this work, we address the problem of offline policy evaluation for manipulation policies under sparse, episode-level rewards, where finite-horizon [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Visualization of the bootstrap mechanism introduced in Sec. V-C. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Success, failure, and composite metrics for each method on the (a) LIBERO-Spatial, (b) Square, and (c) Cloth Folding tasks (dataset size 200 for [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Our method’s predicted values over a successful Square task episode [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Dataset size ablation showing success metric vs. failure metric (higher is better for both) for each method on the (a) LIBERO-Spatial, (b) Square, [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Encoder ablation on the Cloth Folding task (dataset size 150, [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
read the original abstract

Policy evaluation is a fundamental component of the development and deployment pipeline for robotic policies. In modern manipulation systems, this problem is particularly challenging: rewards are often sparse, task progression of evaluation rollouts are often non-monotonic as the policies exhibit recovery behaviors, and evaluation rollouts are necessarily of finite length. This finite length introduces truncation bias, breaking the infinite-horizon assumptions underlying standard methods relying on Bellman equations/principle of optimality. In this work, we propose a framework for offline policy evaluation from sparse rewards based on a liveness-based Bellman operator. Our formulation interprets policy evaluation as a task-completion problem and yields a conservative fixed-point value function that is robust to finite-horizon truncation. We analyze the theoretical properties of the proposed operator, including contraction guarantees, and show how it encodes task progression while mitigating truncation bias. We evaluate our method on two simulated manipulation tasks using both a Vision-Language-Action model and a diffusion policy, and a cloth folding task using human demonstrations. Empirical results demonstrate that our approach more accurately reflects task progress and substantially reduces truncation bias, outperforming classical baselines such as TD(0) and Monte Carlo policy evaluation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a liveness-based Bellman operator for offline policy evaluation of robotic manipulation policies under sparse rewards. It reframes evaluation as a task-completion problem to produce a conservative fixed-point value function that is robust to non-monotonic recovery behaviors and finite-horizon truncation bias. The work analyzes contraction properties of the operator, shows how it encodes task progression without extra labels, and reports empirical outperformance versus TD(0) and Monte Carlo baselines on two simulated tasks (using VLA and diffusion policies) plus a real cloth-folding task with human demonstrations.

Significance. If the contraction analysis and empirical robustness hold, the approach offers a practical advance for policy evaluation in manipulation domains where standard infinite-horizon assumptions fail. Credit is due for the parameter-light formulation (only a liveness discount factor), the explicit handling of non-monotonic progress, and the multi-policy-class evaluation that includes both simulated and real data.

major comments (2)
  1. Abstract and theoretical analysis section: the contraction guarantees and uniqueness of the conservative fixed point are asserted as central results, yet the provided manuscript summary contains no derivation steps, conditions on the liveness discount factor, or proof sketch; this is load-bearing for the claim that the operator remains a contraction under non-monotonic recovery.
  2. Experimental section: the reported outperformance on the real cloth-folding task and reduction in truncation bias are key to the practical contribution, but without details on rollout lengths, how task progress is quantified, or statistical tests against baselines, the strength of the empirical support cannot be fully assessed.
minor comments (2)
  1. Abstract: the phrasing 'task progression of evaluation rollouts are often non-monotonic' contains a subject-verb agreement issue that should be corrected for clarity.
  2. The liveness discount factor is identified as the sole free parameter; a brief sensitivity analysis or default selection guideline would strengthen reproducibility even if not required for the core theory.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback, which helps clarify the presentation of our theoretical and empirical contributions. We address each major comment below.

read point-by-point responses

  1. Referee: Abstract and theoretical analysis section: the contraction guarantees and uniqueness of the conservative fixed point are asserted as central results, yet the provided manuscript summary contains no derivation steps, conditions on the liveness discount factor, or proof sketch; this is load-bearing for the claim that the operator remains a contraction under non-monotonic recovery.

    Authors: The full manuscript contains a dedicated theoretical analysis (Section 3) deriving the contraction properties. To address the concern directly, we will revise the main text to include an explicit outline of the derivation steps, the precise condition that the operator is a contraction for any liveness discount factor γ_l ∈ (0,1), and a short proof sketch showing uniqueness of the conservative fixed point even when recovery behaviors are non-monotonic. The full proof will remain in the appendix. This change makes the central claim self-contained in the main body without altering the underlying results. revision: yes

  2. Referee: Experimental section: the reported outperformance on the real cloth-folding task and reduction in truncation bias are key to the practical contribution, but without details on rollout lengths, how task progress is quantified, or statistical tests against baselines, the strength of the empirical support cannot be fully assessed.

    Authors: We agree that these specifics are needed for rigorous evaluation. In the revised manuscript we will add: rollout lengths (maximum 200 steps in simulation, 100 steps on the real robot), the exact liveness-based quantification of task progress (implicit stage completion without auxiliary labels), and statistical reporting (means, standard deviations over 5 seeds, and paired t-test p-values versus TD(0) and Monte Carlo). These additions will strengthen the evidence for outperformance and truncation-bias reduction. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper defines a new liveness-based Bellman operator from first principles to reinterpret policy evaluation as task completion, then separately proves its contraction mapping property and fixed-point existence. No equation reduces the claimed conservative value function or truncation robustness to a fitted quantity, renamed empirical pattern, or self-citation chain; the derivation chain is self-contained and does not invoke prior results by the same authors as load-bearing uniqueness theorems or ansatzes.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim rests on the novel definition of the liveness-based operator and the domain assumption that task completion can be encoded as a discounted liveness property independent of standard reward accumulation.

free parameters (1)
  • liveness discount factor
    The formulation applies discounting to the liveness signal; this hyperparameter is required to define the operator but its specific value or selection method is not detailed in the abstract.
axioms (1)
  • domain assumption Policy evaluation in manipulation can be reframed as a task-completion liveness problem whose fixed point remains conservative under finite truncation.
    Invoked when interpreting the Bellman operator and claiming robustness to truncation bias.
invented entities (1)
  • Liveness-based Bellman operator no independent evidence
    purpose: To produce a conservative value function that encodes task progression and mitigates truncation bias.
    Newly introduced operator not present in standard RL literature cited.

pith-pipeline@v0.9.0 · 5504 in / 1471 out tokens · 67703 ms · 2026-05-13T02:10:36.026604+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages · 12 internal anchors

  1. [1]

    A new and simpler approximation for ANOV A under variance het- erogeneity.Journal of Educational Statistics, 19(2):91– 101, 1994

    Ralph A Alexander and Diane M Govern. A new and simpler approximation for ANOV A under variance het- erogeneity.Journal of Educational Statistics, 19(2):91– 101, 1994. URL https://www.jstor.org/stable/1165140

    work page arXiv 1994

  2. [2]

    Somil Bansal, Mo Chen, Sylvia Herbert, and Claire J. Tomlin. Hamilton-jacobi reachability: A brief overview and recent advances. In2017 IEEE 56th Annual Con- ference on Decision and Control (CDC), pages 2242– 2253, 2017. doi: 10.1109/CDC.2017.8263977. URL https://arxiv.org/abs/1709.07523

    work page doi:10.1109/cdc.2017.8263977 2017

  3. [3]

    Controlling the false discovery rate: a practical and powerful approach to multiple testing.Journal of the Royal Statistical Society: Series B (Methodological), 57(1):289–300, 1995

    Yoav Benjamini and Yosef Hochberg. Controlling the false discovery rate: a practical and powerful approach to multiple testing.Journal of the Royal Statistical Society: Series B (Methodological), 57(1):289–300, 1995. URL https://www.jstor.org/stable/2346101?seq=1

    work page arXiv 1995

  4. [4]

    GR00T N1: An Open Foundation Model for Generalist Humanoid Robots

    Johan Bjorck, Fernando Casta ˜neda, Nikita Cherniadev, Xingye Da, Runyu Ding, Linxi Fan, Yu Fang, Dieter Fox, Fengyuan Hu, Spencer Huang, et al. Gr00t n1: An open foundation model for generalist humanoid robots.arXiv preprint arXiv:2503.14734, 2025. URL https://arxiv.org/ abs/2503.14734

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  5. [5]

    $\pi_0$: A Vision-Language-Action Flow Model for General Robot Control

    Kevin Black, Noah Brown, Danny Driess, Adnan Es- mail, Michael Equi, Chelsea Finn, Niccolo Fusai, Lachy Groom, Karol Hausman, Brian Ichter, Szymon Jakubczak, Tim Jones, Liyiming Ke, Sergey Levine, Adrian Li-Bell, Mohith Mothukuri, Suraj Nair, Karl Pertsch, Lucy Xiaoyang Shi, James Tanner, Quan Vuong, Anna Walling, Haohuan Wang, and Ury Zhilinsky.π 0: A vi...

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  6. [6]

    RT-1: Robotics Transformer for Real-World Control at Scale

    Anthony Brohan, Noah Brown, Justice Carbajal, Yev- gen Chebotar, Joseph Dabis, Chelsea Finn, Keerthana Gopalakrishnan, Karol Hausman, Alex Herzog, Jasmine Hsu, et al. Rt-1: Robotics transformer for real-world control at scale.arXiv preprint arXiv:2212.06817, 2022. URL https://arxiv.org/abs/2212.06817

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  7. [7]

    Dif- fusion policy: Visuomotor policy learning via action diffusion

    Cheng Chi, Siyuan Feng, Yilun Du, Zhenjia Xu, Eric Cousineau, Benjamin Burchfiel, and Shuran Song. Dif- fusion policy: Visuomotor policy learning via action diffusion. InProceedings of Robotics: Science and Systems (RSS), 2023. URL https://diffusion-policy.cs. columbia.edu/

    work page 2023

  8. [8]

    The convergence oftd(λ)for generalλ

    Peter Dayan. The convergence oftd(λ)for generalλ. Machine Learning, 8(3-4):341–362, 1992. URL https: //link.springer.com/article/10.1007/BF00992701

    work page doi:10.1007/bf00992701 1992

  9. [9]

    Fisac, Neil F

    Jaime F. Fisac, Neil F. Lugovoy, Vicenc Rubies-Royo, Shromona Ghosh, and Claire J. Tomlin. Bridging hamilton-jacobi safety analysis and reinforcement learn- ing. In2019 International Conference on Robotics and Automation (ICRA), pages 8550–8556. IEEE. ISBN 978- 1-5386-6027-0. doi: 10.1109/ICRA.2019.8794107. URL https://ieeexplore.ieee.org/document/8794107/

    work page doi:10.1109/icra.2019.8794107 2019

  10. [10]

    Howard.Dynamic Programming and Markov Processes

    Ronald A. Howard.Dynamic Programming and Markov Processes. The Technology Press of MIT and John Wiley & Sons, Cambridge, MA and New York, NY , 1960. ISBN 9780262080095. URL https://gwern.net/doc/statistics/decision/ 1960-howard-dynamicprogrammingmarkovprocesses. pdf

    work page 1960

  11. [11]

    $\pi^{*}_{0.6}$: a VLA That Learns From Experience

    Physical Intelligence, Ali Amin, Raichelle Aniceto, Ash- win Balakrishna, Kevin Black, Ken Conley, Grace Con- nors, James Darpinian, Karan Dhabalia, Jared DiCarlo, et al.π ∗ 0.6: a vla that learns from experience.arXiv preprint arXiv:2511.14759, 2025. URL https://arxiv.org/ abs/2511.14759

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  12. [12]

    $\pi_{0.5}$: a Vision-Language-Action Model with Open-World Generalization

    Physical Intelligence, Kevin Black, Noah Brown, James Darpinian, Karan Dhabalia, Danny Driess, Adnan Es- mail, Michael Equi, Chelsea Finn, Niccolo Fusai, et al. π0.5: a vision-language-action model with open-world generalization.arXiv preprint arXiv:2504.16054, 2025. URL https://arxiv.org/abs/2504.16054

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  13. [13]

    DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset

    Alexander Khazatsky, Karl Pertsch, Suraj Nair, Ash- win Balakrishna, Sudeep Dasari, Siddharth Karam- cheti, Soroush Nasiriany, Mohan Kumar Srirama, Lawrence Yunliang Chen, Kirsty Ellis, et al. Droid: A large-scale in-the-wild robot manipulation dataset.arXiv preprint arXiv:2403.12945, 2024. URL https://arxiv.org/ abs/2403.12945

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  14. [14]

    LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning

    Bo Liu, Yifeng Zhu, Chongkai Gao, Yihao Feng, Qiang Liu, Yuke Zhu, and Peter Stone. Libero: Benchmarking knowledge transfer for lifelong robot learning.Advances in Neural Information Processing Systems, 36:44776– 44791, 2023. URL https://arxiv.org/abs/2306.03310

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  15. [15]

    On reachability and minimum cost optimal control.Automatica, 40(6):917–927, 2004

    John Lygeros. On reachability and minimum cost optimal control.Automatica, 40(6):917–927, 2004. ISSN 0005-

    work page 2004

  16. [16]

    doi: https://doi.org/10.1016/j.automatica.2004.01

    work page doi:10.1016/j.automatica.2004.01 2004

  17. [17]

    URL https://www.sciencedirect.com/science/article/ pii/S0005109804000263

    work page

  18. [18]

    What Matters in Learning from Offline Human Demonstrations for Robot Manipulation

    Ajay Mandlekar, Danfei Xu, Josiah Wong, Soroush Nasiriany, Chen Wang, Rohun Kulkarni, Li Fei-Fei, Silvio Savarese, Yuke Zhu, and Roberto Mart ´ın-Mart´ın. What matters in learning from offline human demon- strations for robot manipulation. InarXiv preprint arXiv:2108.03298, 2021. URL https://arxiv.org/abs/2108. 03298

    work page internal anchor Pith review arXiv 2021

  19. [19]

    Steering your generalists: Improving robotic foundation models via value guidance.arXiv preprint arXiv:2410.13816, 2024

    Mitsuhiko Nakamoto, Oier Mees, Aviral Kumar, and Sergey Levine. Steering your generalists: Improving robotic foundation models via value guidance.arXiv preprint arXiv:2410.13816, 2024. URL https://arxiv.org/ abs/2410.13816

    work page arXiv 2024

  20. [21]

    URL https://arxiv.org/abs/2304.07193

    work page internal anchor Pith review Pith/arXiv arXiv

  21. [22]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. InInternational conference on machine learning, pages 8748–8763. PmLR, 2021. URL https://proceedings.mlr.press/v139/ radford21a/rad...

    work page 2021

  22. [23]

    Rubinstein and Dirk P

    Reuven Y . Rubinstein and Dirk P. Kroese. Simulation and the Monte Carlo Method. John Wiley & Sons, Hoboken, NJ, 3rd edition,

    work page

  23. [24]

    URL https: //www.wiley.com/en-us/Simulation+and+the+Monte+ Carlo+Method%2C+3rd+Edition-p-9781118632161

    ISBN 978-1-118-63216-1. URL https: //www.wiley.com/en-us/Simulation+and+the+Monte+ Carlo+Method%2C+3rd+Edition-p-9781118632161

    work page

  24. [25]

    Prioritized experience replay

    Tom Schaul, John Quan, Ioannis Antonoglou, and David Silver. Prioritized experience replay.arXiv preprint arXiv:1511.05952, 2015. URL https://arxiv.org/abs/1511. 05952

    work page arXiv 2015

  25. [26]

    Singh and Richard S

    Satinder P. Singh and Richard S. Sutton. Reinforce- ment learning with replacing eligibility traces.Ma- chine Learning, 22(1-3):123–158, 1996. doi: 10.1007/ BF00114726. URL http://www.incompleteideas.net/ papers/singh-sutton-96.ps

    work page 1996

  26. [27]

    Richard S. Sutton. Learning to predict by the methods of temporal differences. 3(1):9–44. ISSN 1573-0565. doi: 10.1007/BF00115009. URL https://doi.org/10.1007/ BF00115009

    work page doi:10.1007/bf00115009

  27. [28]

    MIT press Cambridge, 1998

    Richard S Sutton, Andrew G Barto, et al.Reinforce- ment learning: An introduction, volume 1. MIT press Cambridge, 1998. URL http://incompleteideas.net/book/ the-book-2nd.html

    work page 1998

  28. [29]

    Gemini robotics 1.5: Pushing the frontier of generalist robots with advanced embodied reasoning, thinking, and motion transfer.arXiv preprint arXiv:2510.03342, 2025

    Gemini Robotics Team, Abbas Abdolmaleki, Saminda Abeyruwan, Joshua Ainslie, Jean-Baptiste Alayrac, Montserrat Gonzalez Arenas, Ashwin Balakrishna, Nathan Batchelor, Alex Bewley, Jeff Bingham, et al. Gemini robotics 1.5: Pushing the frontier of generalist robots with advanced embodied reasoning, thinking, and motion transfer.arXiv preprint arXiv:2510.03342...

    work page arXiv 2025

  29. [30]

    Thomas, Georgios Theocharous, and Moham- mad Ghavamzadeh

    Philip S. Thomas, Georgios Theocharous, and Moham- mad Ghavamzadeh. High-confidence off-policy evalua- tion. InProceedings of the Twenty-Ninth AAAI Confer- ence on Artificial Intelligence (AAAI), volume 29, pages 3000–3006, 2015. URL https://ojs.aaai.org/index.php/ AAAI/article/view/9541

    work page 2015

  30. [31]

    SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features

    Michael Tschannen, Alexey Gritsenko, Xiao Wang, Muhammad Ferjad Naeem, Ibrahim Alabdulmohsin, Nikhil Parthasarathy, Talfan Evans, Lucas Beyer, Ye Xia, Basil Mustafa, et al. Siglip 2: Multilingual vision- language encoders with improved semantic understand- ing, localization, and dense features.arXiv preprint arXiv:2502.14786, 2025. URL https://arxiv.org/a...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  31. [32]

    Attention Is All You Need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017. URL https://arxiv.org/abs/1706.03762

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  32. [33]

    Bridgedata v2: A dataset for robot learning at scale, 2024

    Homer Rich Walke, Kevin Black, Tony Z Zhao, Quan Vuong, Chongyi Zheng, Philippe Hansen-Estruch, An- dre Wang He, Vivek Myers, Moo Jin Kim, Max Du, et al. Bridgedata v2: A dataset for robot learning at scale. InConference on Robot Learning, pages 1723–1736. PMLR, 2023. URL https://arxiv.org/abs/2308.12952

    work page arXiv 2023

  33. [34]

    The generalization of ‘Student’s’ problem when several different population variances are involved.Biometrika, 34(1/2):28–35, 1947

    Bernard L Welch. The generalization of ‘Student’s’ problem when several different population variances are involved.Biometrika, 34(1/2):28–35, 1947. URL https: //www.jstor.org/stable/2332510?seq=1

    work page arXiv 1947

  34. [35]

    RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control

    Brianna Zitkovich, Tianhe Yu, Sichun Xu, Peng Xu, Ted Xiao, Fei Xia, Jialin Wu, Paul Wohlhart, Stefan Welker, Ayzaan Wahid, et al. Rt-2: Vision-language- action models transfer web knowledge to robotic control. InConference on Robot Learning, pages 2165–2183. PMLR, 2023. URL https://arxiv.org/abs/2307.15818. APPENDIXA PROOF OFPROPOSITION1 Proof:We use a s...

    work page internal anchor Pith review Pith/arXiv arXiv 2023