Pith. sign in

REVIEW 2 major objections 38 references

Reviewed by Pith at T0; open to challenge.

T0 means a machine referee read the full paper against a public rubric. The mark states how deep the mechanical check went, never who wrote it. the ladder, T0–T4 →

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

T0 review · grok-4.3

Predicting Before/Hit/After tokens lets VLA agents switch tasks and stabilize control with one fixed calibration rule.

2026-06-28 22:36 UTC pith:62CW7P3O

load-bearing objection CaB gives a token-based way to keep two-sided completion evidence for VLA handoffs under a single frozen switching rule, but the abstract supplies no numbers so the actual gains stay hard to judge. the 2 major comments →

arxiv 2606.00145 v1 pith:62CW7P3O submitted 2026-05-29 cs.RO cs.AI

Completion at the Boundary (CaB): Deployable Switching with Completion-Aware Control under Limited Calibration

Yusuke Sano , Takeshi Itoga This is my paper
classification cs.RO cs.AI
keywords vision-language-actiontask completionboundary detectionswitching controlcomposite instructionsMinecraft benchmarklow-calibration deployment
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Vision-language-action agents often fail at short composite instructions because they cannot reliably detect when one sub-task ends and the next must begin. The paper establishes that collapsing boundary evidence into a single scalar is brittle, so it instead outputs three Boundary-Phase Tokens that keep evidence from both sides of the transition. These tokens feed two modules: one decides the exact switch moment and the other conditions ongoing actions to remain stable across the handoff. All of this is required to hold under the strict deployability rule of a single calibration chosen once on development data and never changed at test time.

Core claim

The central claim is that Completion at the Boundary predicts a three-state completion object called Boundary-Phase Tokens (Before/Hit/After) that retains two-sided evidence around each instruction boundary. CaB-When turns this object into an auditable switch decision while CaB-How reuses the same object to condition action generation for boundary-stable control. When evaluated with an intervention-aware E1/E2 protocol on a first-person Minecraft VLA benchmark, the method produces higher composite execution success and better handoff quality than scalar baselines under matched model capacity and the low-calibration constraint of one unchanging global rule.

What carries the argument

Boundary-Phase Tokens, a three-state prediction (Before/Hit/After) that supplies two-sided boundary evidence for both the switching decision and the action-conditioning step.

Load-bearing premise

A single switching rule chosen once on development data remains effective for open-ended instructions while the three-phase tokens still supply usable evidence on both sides of each boundary.

What would settle it

On the Minecraft VLA benchmark, run the same E1/E2 protocol with CaB replaced by a scalar completion predictor that uses the identical single global calibration rule; if composite success rate and handoff quality show no gain or a loss, the central claim is false.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Composite task success rises because switches occur at better-timed moments.
  • Handoff quality improves because action generation is conditioned on the three-phase evidence through the transition.
  • The same completion object serves both the when decision and the how control without extra modules.
  • All gains hold under the constraint of no test-time relearning and one fixed calibration rule.

Where Pith is reading between the lines

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

  • The same three-phase object could be attached to existing VLA models to reduce error accumulation across longer instruction chains.
  • Retaining asymmetric evidence may help in any sequential decision setting where polarity of observations changes across a boundary.
  • If the tokens prove stable, downstream planners could treat the Hit state as an explicit synchronization point rather than an implicit switch.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 0 minor

Summary. The manuscript proposes Completion at the Boundary (CaB) for vision-language-action (VLA) agents to decide instruction completion in short composites under a deployable low-calibration regime (no test-time relearning, single globally calibrated switching rule chosen once on development set). It introduces Boundary-Phase Tokens (Before/Hit/After) to retain two-sided boundary evidence, with CaB-When converting the object to a switching decision and CaB-How reusing it to condition actions for boundary-stable control. Using an intervention-aware E1/E2 protocol, the paper claims improvements in composite execution and handoff quality on a first-person Minecraft VLA benchmark under matched capacity and deployability constraints.

Significance. If the results hold, the work addresses an operational gap in deployed VLA systems for open-ended instructions by enabling reliable handoffs without per-task recalibration. The low-calibration discipline and intervention-aware E1/E2 protocol are practical strengths that could support reproducible evaluation in robotics.

major comments (2)
  1. [Abstract] Abstract: the central empirical claim of improved composite execution and handoff quality is stated without any quantitative results, baselines, error bars, or metric definitions, so the magnitude and attribution of gains cannot be assessed from the text.
  2. [Abstract] Abstract (and sections describing the low-calibration regime and CaB design): the assertion that Boundary-Phase Tokens retain usable two-sided evidence under arbitrary task polarity shifts with a single frozen switching rule is not supported by any derivation, ablation, or empirical breakdown; this is load-bearing for attributing E1/E2 gains to CaB rather than the protocol itself.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and commit to revisions that improve clarity and attribution without altering the core claims or experimental design.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central empirical claim of improved composite execution and handoff quality is stated without any quantitative results, baselines, error bars, or metric definitions, so the magnitude and attribution of gains cannot be assessed from the text.

    Authors: We agree that the abstract would benefit from explicit quantitative anchors. In the revised version we will insert concise results (e.g., composite success rate deltas, handoff-quality scores, baseline comparisons, and standard-error ranges) together with one-sentence metric definitions, while preserving the abstract length limit. revision: yes

  2. Referee: [Abstract] Abstract (and sections describing the low-calibration regime and CaB design): the assertion that Boundary-Phase Tokens retain usable two-sided evidence under arbitrary task polarity shifts with a single frozen switching rule is not supported by any derivation, ablation, or empirical breakdown; this is load-bearing for attributing E1/E2 gains to CaB rather than the protocol itself.

    Authors: The intervention-aware E1/E2 protocol already enforces the single frozen rule across the test distribution, and the reported gains are measured under that constraint. Nevertheless, we accept that an explicit polarity-shift ablation would strengthen the causal link to the two-sided Boundary-Phase Tokens. We will add a targeted breakdown (performance stratified by task polarity) and a short derivation sketch of why the three-token representation preserves evidence under sign flips; these additions will appear in the main text and appendix. revision: yes

Circularity Check

0 steps flagged

No circularity detected; derivation chain is self-contained

full rationale

The provided abstract and description contain no equations, fitted parameters presented as predictions, self-citations, or ansatzes that reduce any claim to its own inputs by construction. The low-calibration regime and Boundary-Phase Tokens are introduced as design choices under explicit constraints, with empirical results on a benchmark; no load-bearing step matches any enumerated circularity pattern. This is the expected non-finding for an empirical methods paper without mathematical derivation.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

Abstract introduces Boundary-Phase Tokens and the low-calibration regime without enumerating free parameters, background axioms, or independent evidence for the new tokens.

invented entities (1)
  • Boundary-Phase Tokens (Before/Hit/After) no independent evidence
    purpose: Retain two-sided boundary evidence for completion prediction under single global calibration
    New representation introduced to avoid brittleness of scalar collapse; no independent evidence supplied in abstract

pith-pipeline@v0.9.1-grok · 5771 in / 1322 out tokens · 23898 ms · 2026-06-28T22:36:47.730827+00:00 · methodology

0 comments
read the original abstract

Vision-language-action (VLA) agents can execute natural-language instructions, yet deployed systems still lack an operational interface: deciding when the instruction is complete. This gap is acute in short composites ("do A, then B"), where mistimed handoffs cascade into downstream failures. Completion is inherently closed-loop because switching is an intervention that changes the instruction context and thus future actions and observations. We study completion under a deployable low-calibration regime motivated by open-ended instruction spaces, enforcing no test-time relearning and a single globally calibrated switching rule selected once on development set and reused unchanged on test set. Under this constraint, collapsing asymmetric boundary evidence into a single scalar can be brittle under polarity shifts across tasks. We propose Completion at the Boundary (CaB), which predicts an event-local completion object in the form of Boundary-Phase Tokens (Before/Hit/After), retaining two-sided boundary evidence under this discipline. CaB-When converts this completion object into a minimal, auditable switching decision (when), while CaB-How reuses the same completion object to condition action generation for boundary-stable control through handoffs (how). Using an intervention-aware E1/E2 protocol, we show that CaB improves composite execution and handoff quality on a first-person Minecraft VLA benchmark under matched capacity and deployability constraints.

Figures

Figures reproduced from arXiv: 2606.00145 by Takeshi Itoga, Yusuke Sano.

Figure 1
Figure 1. Figure 1: Key idea: deployable completion interface under intervention. (a) In short composite instructions (“do A, then B”), mistimed handoffs (premature vs. overrun) cause downstream failures; switching updates the active instruction and thus intervenes in the trajectory. (b) Many VLA policies lack an explicit completion interface for deciding when to switch, whereas CaB learns an event-local boundary-phase poster… view at source ↗
Figure 2
Figure 2. Figure 2: CaB overview (one posterior, two consumers). An autoregressive VLA policy jointly predicts actions and an event-local BPT posterior pt(y) as the shared completion object. This poste￾rior is dual-used by two consumers: (i) CaB-When applies a fixed readout to decide when to switch, and (ii) CaB-How reuses it to condition how actions are generated for boundary-stable control. 3 Problem Setting We consider dis… view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

38 extracted references · 12 canonical work pages · 2 internal anchors

  1. [1]

    Hind- sight experience replay

    Marcin Andrychowicz, Dwight Crow, Alex Ray, Jonas Schneider, Rachel Fong, Peter Welinder, Bob McGrew, Josh Tobin, Pieter Abbeel, and Wojciech Zaremba. Hind- sight experience replay. In Isabelle Guyon, Ulrike von Luxburg, Samy Bengio, Hanna M. Wallach, Rob Fergus, S. V . N. Vishwanathan, and Roman Garnett, editors, Advances in Neural Information Processing...

    2017

  2. [3]

    The option-critic architecture

    Pierre-Luc Bacon, Jean Harb, and Doina Precup. The option-critic architecture. In Satinder Singh and Shaul Markovitch, editors,Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence, February 4-9, 2017, San Francisco, California, USA, pages 1726–1734. AAAI Press, 2017. doi: 10.1609/AAAI.V31I1.10916. URLhttps://doi.org/10.1609/ aaai.v31i1.10916

    work page doi:10.1609/aaai.v31i1.10916 2017

  3. [4]

    Video pretraining (VPT): learning to act by watching unlabeled online videos

    Bowen Baker, Ilge Akkaya, Peter Zhokhov, Joost Huizinga, Jie Tang, Adrien Ecof- fet, Brandon Houghton, Raul Sampedro, and Jeff Clune. Video pretraining (VPT): learning to act by watching unlabeled online videos. In Sanmi Koyejo, S. Mohamed, A. Agarwal, Danielle Belgrave, K. Cho, and A. Oh, editors,Advances in Neural In- formation Processing Systems 35: An...

    2022

  4. [6]

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

    Kevin Black, Noah Brown, Danny Driess, Adnan Esmail, 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 Xi- aoyang Shi, James Tanner, Quan Vuong, Anna Walling, Haohuan Wang, and Ury Zhilinsky. π0: A vi...

    work page internal anchor Pith review Pith/arXiv arXiv

  5. [8]

    Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, Joseph Dabis, Chelsea Finn, Keerthana Gopalakrishnan, Karol Hausman, Alexander Herzog, Jasmine Hsu, Julian Ibarz, Brian Ichter, Alex Irpan, Tomas Jackson, Sally Jesmonth, Nikhil J. Joshi, Ryan Julian, Dmitry Kalashnikov, Yuheng Kuang, Isabel Leal, Kuang-Huei Lee, Sergey Levine, Yao Lu, Ut- sav...

    work page doi:10.15607/rss.2023.xix.025 2023

  6. [10]

    Decision transformer: Rein- forcement learning via sequence modeling

    Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, and Igor Mordatch. Decision transformer: Rein- forcement learning via sequence modeling. In Marc’Aurelio Ranzato, Alina Beygelz- imer, Yann N. Dauphin, Percy Liang, and Jennifer Wortman Vaughan, editors,Ad- vances in Neural Information Proce...

    2021

  7. [11]

    Regression models and life-tables.Journal of the royal statistical society: Series B (methodological), 34(2):187–202, 1972

    David R Cox. Regression models and life-tables.Journal of the royal statistical society: Series B (methodological), 34(2):187–202, 1972

    1972

  8. [13]

    Trivedi, Utkarsh Upadhyay, Manuel Gomez-Rodriguez, and Le Song

    Nan Du, Hanjun Dai, Rakshit S. Trivedi, Utkarsh Upadhyay, Manuel Gomez-Rodriguez, and Le Song. Recurrent marked temporal point processes: Embedding event history to vector. In Balaji Krishnapuram, Mohak Shah, Alexander J. Smola, Charu C. Aggarwal, Dou Shen, and Rajeev Rastogi, editors,Proceedings of the 22nd ACM SIGKDD International Conference on Knowledg...

    work page doi:10.1145/2939672.2939875 2016

  9. [14]

    Vision-language models as success detectors

    Yuqing Du, Ksenia Konyushkova, Misha Denil, Akhil Raju, Jessica Landon, Felix Hill, Nando de Freitas, and Serkan Cabi. Vision-language models as success detectors. In Sarath Chandar, Razvan Pascanu, Hanie Sedghi, and Doina Precup, editors,Conference on Lifelong Learning Agents, 22-25 August 2023, McGill University, Montr ´eal, Qu ´ebec, Canada, Proceeding...

    2023

  10. [15]

    Rvs: What is essen- tial for offline RL via supervised learning? InThe Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022

    Scott Emmons, Benjamin Eysenbach, Ilya Kostrikov, and Sergey Levine. Rvs: What is essen- tial for offline RL via supervised learning? InThe Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022. OpenReview.net, 2022. URL https://openreview.net/forum?id=S874XAIpkR-

    2022

  11. [16]

    Minedojo: Building open-ended embodied agents with internet-scale knowledge

    Linxi Fan, Guanzhi Wang, Yunfan Jiang, Ajay Mandlekar, Yuncong Yang, Haoyi Zhu, Andrew Tang, De-An Huang, Yuke Zhu, and Anima Anandkumar. Minedojo: Building open-ended embodied agents with internet-scale knowledge. In Sanmi Koyejo, S. Mo- hamed, A. Agarwal, Danielle Belgrave, K. Cho, and A. Oh, editors,Advances in Neu- ral Information Processing Systems 3...

    2022

  12. [17]

    Self-improving embodied foundation models.arXiv preprint arXiv:2509.15155, 2025

    Seyed Kamyar Seyed Ghasemipour, Ayzaan Wahid, Jonathan Tompson, Pannag Sanketi, and Igor Mordatch. Self-improving embodied foundation models.CoRR, abs/2509.15155,

    work page arXiv

  13. [20]

    Guss, Brandon Houghton, Nicholay Topin, Phillip Wang, Cayden R

    William H. Guss, Brandon Houghton, Nicholay Topin, Phillip Wang, Cayden R. Codel, Manuela Veloso, and Ruslan Salakhutdinov. Minerl: A large-scale dataset of minecraft demon- strations. In Sarit Kraus, editor,Proceedings of the Twenty-Eighth International Joint Confer- ence on Artificial Intelligence, IJCAI 2019, Macao, China, August 10-16, 2019, pages 2442–

    2019

  14. [21]

    doi: 10.24963/IJCAI.2019/339

    ijcai.org, 2019. doi: 10.24963/IJCAI.2019/339. URLhttps://doi.org/10.24963/ ijcai.2019/339

    work page doi:10.24963/ijcai.2019/339 2019

  15. [22]

    Joshi, Kyle Jeffrey, Rosario Jauregui Ruano, Jasmine Hsu, Keerthana Gopalakrishnan, Byron David, Andy Zeng, and Chuyuan Kelly Fu

    Brian Ichter, Anthony Brohan, Yevgen Chebotar, Chelsea Finn, Karol Hausman, Alexander Herzog, Daniel Ho, Julian Ibarz, Alex Irpan, Eric Jang, Ryan Julian, Dmitry Kalashnikov, Sergey Levine, Yao Lu, Carolina Parada, Kanishka Rao, Pierre Sermanet, Alexander Toshev, Vincent Vanhoucke, Fei Xia, Ted Xiao, Peng Xu, Mengyuan Yan, Noah Brown, Michael Ahn, Omar Co...

    2022

  16. [25]

    Prismatic vlms: Investigating the design space of visually-conditioned lan- guage models

    Siddharth Karamcheti, Suraj Nair, Ashwin Balakrishna, Percy Liang, Thomas Kollar, and Dorsa Sadigh. Prismatic vlms: Investigating the design space of visually-conditioned lan- guage models. In Ruslan Salakhutdinov, Zico Kolter, Katherine A. Heller, Adrian Weller, Nuria Oliver, Jonathan Scarlett, and Felix Berkenkamp, editors,Forty-first International Con-...

    2024

  17. [26]

    Sanketi, Quan Vuong, Thomas Kollar, Ben- jamin Burchfiel, Russ Tedrake, Dorsa Sadigh, Sergey Levine, Percy Liang, and Chelsea Finn

    Moo Jin Kim, Karl Pertsch, Siddharth Karamcheti, Ted Xiao, Ashwin Balakrishna, Suraj Nair, Rafael Rafailov, Ethan Paul Foster, Pannag R. Sanketi, Quan Vuong, Thomas Kollar, Ben- jamin Burchfiel, Russ Tedrake, Dorsa Sadigh, Sergey Levine, Percy Liang, and Chelsea Finn. Openvla: An open-source vision-language-action model. In Pulkit Agrawal, Oliver Kroemer,...

    2024

  18. [27]

    JARVIS-VLA: post- training large-scale vision language models to play visual games with keyboards and mouse

    Muyao Li, Zihao Wang, Kaichen He, Xiaojian Ma, and Yitao Liang. JARVIS-VLA: post- training large-scale vision language models to play visual games with keyboards and mouse. 12 In Wanxiang Che, Joyce Nabende, Ekaterina Shutova, and Mohammad Taher Pilehvar, editors, Findings of the Association for Computational Linguistics, ACL 2025, Vienna, Austria, July 2...

    2025

  19. [29]

    Gupta, Rilyn Han, Li Fei-Fei, and Saining Xie

    Zaijing Li, Yuquan Xie, Rui Shao, Gongwei Chen, Dongmei Jiang, and Liqiang Nie. Optimus-2: Multimodal minecraft agent with goal-observation-action conditioned policy. InIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2025, Nashville, TN, USA, June 11-15, 2025, pages 9039–9049. Computer Vision Foundation / IEEE, 2025. doi: 10.1109/CVPR...

    work page doi:10.1109/cvpr52734.2025.00845 2025

  20. [30]

    McIlraith

    Shalev Lifshitz, Keiran Paster, Harris Chan, Jimmy Ba, and Sheila A. McIlraith. STEVE- 1: A generative model for text-to-behavior in minecraft. In Alice Oh, Tristan Nau- mann, Amir Globerson, Kate Saenko, Moritz Hardt, and Sergey Levine, editors,Ad- vances in Neural Information Processing Systems 36: Annual Conference on Neural In- formation Processing Sy...

    2023

  21. [32]

    Self-monitoring navigation agent via auxiliary progress estimation

    Chih-Yao Ma, Jiasen Lu, Zuxuan Wu, Ghassan AlRegib, Zsolt Kira, Richard Socher, and Caiming Xiong. Self-monitoring navigation agent via auxiliary progress estimation. In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019. OpenReview.net, 2019. URLhttps://openreview.net/forum?id= r1GAsjC5Fm

    2019

  22. [33]

    Vision language models are in-context value learners

    Yecheng Jason Ma, Joey Hejna, Chuyuan Fu, Dhruv Shah, Jacky Liang, Zhuo Xu, Sean Kir- mani, Peng Xu, Danny Driess, Ted Xiao, Osbert Bastani, Dinesh Jayaraman, Wenhao Yu, Tingnan Zhang, Dorsa Sadigh, and Fei Xia. Vision language models are in-context value learners. InThe Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore...

    2025

  23. [34]

    The neural hawkes process: A neurally self-modulating multivariate point process

    Hongyuan Mei and Jason Eisner. The neural hawkes process: A neurally self-modulating multivariate point process. In Isabelle Guyon, Ulrike von Luxburg, Samy Bengio, Hanna M. Wallach, Rob Fergus, S. V . N. Vishwanathan, and Roman Garnett, editors, Advances in Neural Information Processing Systems 30: Annual Conference on Neu- ral Information Processing Sys...

    2017

  24. [35]

    Visual reinforcement learning with imagined goals

    Ashvin Nair, Vitchyr Pong, Murtaza Dalal, Shikhar Bahl, Steven Lin, and Sergey Levine. Visual reinforcement learning with imagined goals. In Samy Bengio, Hanna M. Wallach, Hugo Larochelle, Kristen Grauman, Nicol `o Cesa-Bianchi, and Roman Garnett, editors,Ad- vances in Neural Information Processing Systems 31: Annual Conference on Neural Infor- mation Pro...

    2018

  25. [36]

    Temporal difference models: Model-free deep RL for model-based control

    Vitchyr Pong, Shixiang Gu, Murtaza Dalal, and Sergey Levine. Temporal difference models: Model-free deep RL for model-based control. In6th International Conference on Learning Representations, ICLR 2018, Vancouver, BC, Canada, April 30 - May 3, 2018, Conference Track Proceedings. OpenReview.net, 2018. URLhttps://openreview.net/forum?id= Skw0n-W0Z

    2018

  26. [37]

    Universal value function ap- proximators

    Tom Schaul, Daniel Horgan, Karol Gregor, and David Silver. Universal value function ap- proximators. In Francis R. Bach and David M. Blei, editors,Proceedings of the 32nd In- ternational Conference on Machine Learning, ICML 2015, Lille, France, 6-11 July 2015, JMLR Workshop and Conference Proceedings, pages 1312–1320. JMLR.org, 2015. URL http://proceeding...

    2015

  27. [38]

    Adafactor: Adaptive learning rates with sublinear memory cost

    Noam Shazeer and Mitchell Stern. Adafactor: Adaptive learning rates with sublinear memory cost. In Jennifer G. Dy and Andreas Krause, editors,Proceedings of the 35th International Conference on Machine Learning, ICML 2018, Stockholmsm ¨assan, Stockholm, Sweden, July 10-15, 2018, Proceedings of Machine Learning Research, pages 4603–4611. PMLR, 2018. URLhtt...

    2018

  28. [39]

    Hi robot: Open-ended instruction following with hierarchical vision-language-action models

    Lucy Xiaoyang Shi, Brian Ichter, Michael Robert Equi, Liyiming Ke, Karl Pertsch, Quan Vuong, James Tanner, Anna Walling, Haohuan Wang, Niccolo Fusai, Adrian Li-Bell, Danny Driess, Lachy Groom, Sergey Levine, and Chelsea Finn. Hi robot: Open-ended instruction following with hierarchical vision-language-action models. In Aarti Singh, Maryam Fazel, Daniel Hs...

    2025

  29. [41]

    Sutton, Doina Precup, and Satinder Singh

    Richard S. Sutton, Doina Precup, and Satinder Singh. Between mdps and semi-mdps: A framework for temporal abstraction in reinforcement learning.Artif. Intell., 112(1-2):181– 211, 1999. doi: 10.1016/S0004-3702(99)00052-1. URLhttps://doi.org/10.1016/ S0004-3702(99)00052-1

    work page doi:10.1016/s0004-3702(99)00052-1 1999

  30. [42]

    V oyager: An open-ended embodied agent with large language models.Trans

    Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. V oyager: An open-ended embodied agent with large language models.Trans. Mach. Learn. Res., 2024, 2024. URLhttps://openreview.net/forum? id=ehfRiF0R3a

    2024

  31. [43]

    Omnijarvis: Unified vision-language-action tokenization enables open-world instruction following agents

    Zihao Wang, Shaofei Cai, Zhancun Mu, Haowei Lin, Ceyao Zhang, Xuejie Liu, Qing Li, Anji Liu, Xiaojian (Shawn) Ma, and Yitao Liang. Omnijarvis: Unified vision-language-action tokenization enables open-world instruction following agents. In Amir Globersons, Lester Mackey, Danielle Belgrave, Angela Fan, Ulrich Paquet, Jakub M. Tomczak, and Cheng Zhang, edito...

    2024

  32. [47]

    Agentic Robot: A Brain-Inspired Framework for Vision-Language-Action Models in Embodied Agents

    Zhejian Yang, Yongchao Chen, Xueyang Zhou, Jiangyue Yan, Dingjie Song, Yinuo Liu, Yut- ing Li, Yu Zhang, Pan Zhou, Hechang Chen, and Lichao Sun. Agentic robot: A brain-inspired framework for vision-language-action models in embodied agents.CoRR, abs/2505.23450,

    work page Pith review arXiv

  33. [49]

    2509.13688

    Haoqi Yuan, Chi Zhang, Hongcheng Wang, Feiyang Xie, Penglin Cai, Hao Dong, and Zongqing Lu. Plan4mc: Skill reinforcement learning and planning for open-world minecraft tasks.CoRR, abs/2303.16563, 2023. doi: 10.48550/ARXIV .2303.16563. URLhttps: //doi.org/10.48550/arXiv.2303.16563

    work page internal anchor Pith review doi:10.48550/arxiv 2023

  34. [50]

    A vision-language-action-critic model for robotic real-world reinforcement learning

    Shaopeng Zhai, Qi Zhang, Tianyi Zhang, Fuxian Huang, Haoran Zhang, Ming Zhou, Shengzhe Zhang, Litao Liu, Sixu Lin, and Jiangmiao Pang. A vision-language-action-critic model for robotic real-world reinforcement learning.CoRR, abs/2509.15937, 2025. doi: 10.48550/ ARXIV .2509.15937. URLhttps://doi.org/10.48550/arXiv.2509.15937

    work page doi:10.48550/arxiv.2509.15937 2025

  35. [51]

    Pytorch FSDP: experiences on scaling fully sharded data parallel.Proc

    Yanli Zhao, Andrew Gu, Rohan Varma, Liang Luo, Chien-Chin Huang, Min Xu, Less Wright, Hamid Shojanazeri, Myle Ott, Sam Shleifer, Alban Desmaison, Can Balioglu, Pritam Dama- nia, Bernard Nguyen, Geeta Chauhan, Yuchen Hao, Ajit Mathews, and Shen Li. Pytorch FSDP: experiences on scaling fully sharded data parallel.Proc. VLDB Endow., 16(12):3848–3860,

  36. [52]

    doi: 10.14778/3611540.3611569.https: //doi.org/10.14778/3611540.3611569

    doi: 10.14778/3611540.3611569. URLhttps://www.vldb.org/pvldb/vol16/ p3848-huang.pdf

    work page doi:10.14778/3611540.3611569

  37. [53]

    Pan, Zhangyang Wang, and Jinwon Lee

    Hanqing Zhu, Zhenyu Zhang, Wenyan Cong, Xi Liu, Sem Park, Vikas Chandra, Bo Long, David Z. Pan, Zhangyang Wang, and Jinwon Lee. APOLLO: sgd-like memory, adamw-level performance. In Matei Zaharia, Gauri Joshi, and Yingyan (Celine) Lin, editors,Proceedings of the Eighth Conference on Machine Learning and Systems, MLSys 2025, Santa Clara, CA, USA, May 12-15,...

    2025

  38. [54]

    fixed wrapper + one global rule(θ, L)

    Brianna Zitkovich, Tianhe Yu, Sichun Xu, Peng Xu, Ted Xiao, Fei Xia, Jialin Wu, Paul Wohlhart, Stefan Welker, Ayzaan Wahid, Quan Vuong, Vincent Vanhoucke, Huong T. Tran, Radu Soricut, Anikait Singh, Jaspiar Singh, Pierre Sermanet, Pannag R. Sanketi, Grecia Salazar, Michael S. Ryoo, Krista Reymann, Kanishka Rao, Karl Pertsch, Igor Mordatch, Henryk Michalew...

    2023