Pith. sign in

REVIEW 2 major objections 61 references

On-Device Robotic Planning: Eliminating Inference Redundancy for Efficient Decision-Making

T0 review · 2 major / 0 minor · reviewed 2026-06-28 · grok-4.3

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

Pith's one-line read Robotic reasoning can skip most inferences when consecutive observations repeat the same actions and subgoals.

desk verdict REIS targets a real latency issue in LLM robot planners by skipping repeated inferences, but the key redundancy assumption gets no numbers or independent checks in the visible material. read the letter →

arxiv 2605.31460 v1 pith:EUBXDN46 submitted 2026-05-29 cs.RO cs.SYeess.SY

Joonhee Lee , Hyunseung Shin , Hyunmi Kim , Pei Zhang , Jeonggil Ko This is my paper · ORCID
classification cs.ROcs.SYeess.SY
keywords roboticplanninginferenceoptimizationtemporalredundancyon-devicedeploymentvision-languagemodelsALFREDbenchmarkscenegating
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 shows that large-model robotic planners waste compute on repeated decisions across time steps. It introduces REIS to detect and bypass this redundancy with lightweight checks before full reasoning. The method keeps task success rates competitive on ALFRED and real robots while cutting inference load. A sympathetic reader would see this as a practical route to real-time on-device control without retraining the underlying models.

What carries the argument

REIS framework, which uses lightweight scene gating to detect redundancy, KV-steered affordance routing to select routes, and deliberative reasoning only when needed.

What would settle it

Measure the fraction of consecutive time steps that produce identical actions and subgoals in a new robotic environment or model; if the fraction is near zero, the claimed overhead reduction disappears.

Watch

Extended reading notes

Core claim

Robotic reasoning workloads contain substantial temporal redundancy where consecutive observations frequently produce identical actions and subgoals; REIS exploits this by combining lightweight scene gating, KV-steered affordance routing, and deliberative reasoning to minimize unnecessary inference while preserving semantic adaptability and competitive performance on ALFRED and real-world tasks.

Load-bearing premise

Robotic tasks produce enough repeated actions and subgoals across consecutive observations that skipping them does not hurt overall performance.

Editorial extensions

If this is right

  • On-device robotic policies become feasible with lower latency under embodied constraints.
  • Task completion rates on ALFRED and physical robots stay close to full-reasoning baselines.
  • The same redundancy pattern can be exploited in other sequential embodied tasks without changing the base model.

Reading between the lines

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

  • If temporal redundancy holds across a wider range of vision-language models, the approach could generalize beyond the tested planners.
  • Real-world deployment would still require verifying that the gating mechanism itself stays cheap on target hardware.
  • The same observation of repeated decisions might apply to non-robotic sequential planning problems such as game agents or autonomous vehicles.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Request a human review

A listed scientist reviews the paper for a fee and the review publishes here regardless of verdict. See the reviewers or get listed.

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 paper claims that robotic reasoning workloads exhibit substantial temporal redundancy (consecutive observations frequently yielding identical actions and subgoals), and introduces the REIS framework—combining lightweight scene gating, KV-steered affordance routing, and deliberative reasoning—to suppress inference overhead while preserving semantic adaptability and competitive task performance on ALFRED and real-world robotic tasks.

Significance. If the redundancy observation holds with independent quantification and the components deliver the claimed latency reductions without hidden costs to adaptability, the work would address a key barrier to real-time on-device deployment of VLM/LLM-based robotic planners, offering a practical path to lower-latency embodied decision-making.

major comments (2)
  1. [Abstract] Abstract: the central claim that REIS 'significantly suppresses reasoning overhead' rests on the unquantified observation of 'substantial temporal redundancy'; no metrics (e.g., fraction of identical consecutive actions/subgoals across ALFRED trajectories or robot logs, or ablation across other VLMs) are reported, so it is impossible to verify whether the property is general or setup-specific.
  2. [Abstract] Abstract: without baseline comparisons, error bars, or explicit measurement of redundancy and overhead (e.g., tokens or latency before/after each REIS component), the performance claims cannot be assessed for soundness or generality.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the need to strengthen the abstract with explicit quantification. We address each major comment below and will revise the manuscript to incorporate the requested metrics and clarifications.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that REIS 'significantly suppresses reasoning overhead' rests on the unquantified observation of 'substantial temporal redundancy'; no metrics (e.g., fraction of identical consecutive actions/subgoals across ALFRED trajectories or robot logs, or ablation across other VLMs) are reported, so it is impossible to verify whether the property is general or setup-specific.

    Authors: We agree that the abstract would benefit from explicit quantification of the temporal redundancy observation. The full manuscript provides supporting experimental analysis on ALFRED and real-robot trajectories, but we will revise the abstract (and add a dedicated quantification subsection) to report concrete metrics such as the fraction of identical consecutive actions and subgoals. Ablations across additional VLMs will be included where existing data permits; new experiments on further models are not feasible within the revision timeline but can be noted as future work. revision: yes

  2. Referee: [Abstract] Abstract: without baseline comparisons, error bars, or explicit measurement of redundancy and overhead (e.g., tokens or latency before/after each REIS component), the performance claims cannot be assessed for soundness or generality.

    Authors: The experimental section already contains baseline comparisons, latency/token measurements, and component-wise breakdowns. To directly address the abstract-level concern, we will revise the abstract to include key quantitative results (with error bars) and explicit before/after overhead numbers for each REIS component (scene gating, KV-steered routing, deliberative reasoning). This will make the performance claims immediately verifiable from the abstract. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity; framework rests on empirical observation without self-referential derivations

full rationale

The paper states an observation about temporal redundancy in robotic reasoning workloads and introduces REIS components (scene gating, KV-steered routing, deliberative reasoning) as engineering responses, validated empirically on ALFRED and real-robot tasks. No equations, parameter-fitting steps, self-citations, or uniqueness theorems appear in the provided text. The central claim of overhead suppression follows from experimental results rather than any reduction of outputs to inputs by construction, making the derivation self-contained against external benchmarks.

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

Abstract provides no equations, parameters, or explicit assumptions beyond the stated observation of temporal redundancy; no free parameters, axioms, or invented entities can be extracted.

how reviews work

0 comments
Cite this review

Pith. "Pith review of On-Device Robotic Planning: Eliminating Inference Redundancy for Efficient Decision-Making." pith.science (2026). https://pith.science/paper/EUBXDN46

@misc{pith2026260531460,
  author       = {Pith},
  title        = {Pith review of: On-Device Robotic Planning: Eliminating Inference Redundancy for Efficient Decision-Making},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EUBXDN46}},
  note         = {Machine review of arXiv:2605.31460}
}
read the original abstract

Reasoning-based robotic policies using large language and vision-language models achieve strong semantic planning capabilities but mostly suffer from a high inference latency that limits practical real-time deployment. In this work, we observe that robotic reasoning workloads contain substantial temporal redundancy, where consecutive observations frequently produce identical actions and subgoals. Based on this insight, we present REIS, a human cognition inspired robotic decision-making framework that minimizes unnecessary reasoning while preserving semantic adaptability. REIS combines lightweight scene gating, KV-steered affordance routing, and deliberative reasoning to accelerate robotic control under embodied constraints. Experiments on ALFRED, and real-world robotic tasks demonstrate that REIS significantly suppresses reasoning overhead while maintaining competitive task performance.

Figures

Figures reproduced from arXiv: 2605.31460 by the authors.

Figure 1
Figure 1. Overview of REIS, illustrating the interaction between System One and System Two. Detailed illustration of System One and Two on the bottom row substantial structural scene changes are detected. Rather than using all visual features, EMA-HSVS selects transformer heads sensitive to geometric variation through a greedy calibration process on datasets such as ALFRED and LIBERO. For incoming frames, EMA-HSVS first trims… view at source ↗
Figure 2
Figure 2. Example of Datasets used for Optimization and Evaluation [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. A Sample of Navigation Task (DNF: Did Not Finish) [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (8 more)
Figure 5
Figure 5. Figure 5: Confusion Matrix of Vision Similarity Measure Methods [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Pipeline of Head Selection (Figure 3 from Main Text) [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: Visual pipelines for (Left) Contrastive Pair Generation (Figure 4) and (Right) Calculation [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: Positive and Negative Prompt for KV Steering Pair Generation [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: Sweep Result of EMA-HSVS Greedy Search [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]
Figure 10
Figure 10. Figure 10: Hallway with Bright Color and Complex Objects on The Wall [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]
Figure 11
Figure 11. Figure 11: Hallway with Monotone Color and Dynamic Pathway Appearing [PITH_FULL_IMAGE:figures/full_fig_p018_11.png]
Figure 12
Figure 12. Figure 12: Comparison of Other Transformer Based Vision Encoder Performance [PITH_FULL_IMAGE:figures/full_fig_p018_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

61 extracted references · 34 canonical work pages

  1. [1]

    Q. Zhao, Y . Lu, M. J. Kim, Z. Fu, Z. Zhang, Y . Wu, Z. Li, Q. Ma, S. Han, C. Finn, A. Handa, M.-Y . Liu, D. Xiang, G. Wetzstein, and T.-Y . Lin. Cot-vla: Visual chain-of-thought reasoning for vision-language-action models. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 2025. URLhttps://openaccess.thecvf.com/ conten...

    2025

  2. [2]

    J. Wen, Y . Zhu, M. Zhu, Z. Tang, J. Li, Z. Zhou, X. Liu, C. Shen, Y . Peng, and F. Feng. DiffusionVLA: Scaling robot foundation models via unified diffusion and autoregression. InProceedings of the 42nd International Conference on Machine Learning, volume 267 ofProceedings of Machine Learning Research, pages 66558–66574. PMLR, 2025. URL https://proceedin...

    2025

  3. [3]

    2026 , eprint =

    W. Jiang, J. Clemons, K. Sankaralingam, and C. Kozyrakis. How fast can i run my vla? demystifying vla inference performance with vla-perf, 2026. URLhttps://arxiv.org/ abs/2602.18397

    work page arXiv 2026

  4. [4]

    Y . Dai, H. Gu, T. Wang, Q. Cheng, Y . Zheng, Z. Qiu, L. Gong, W. Lou, and X. Zhou. Ac- tionflow: A pipelined action acceleration for vision language models on edge.arXiv preprint arXiv:2512.20276, 2025. URLhttps://arxiv.org/abs/2512.20276

    work page arXiv 2025

  5. [5]

    Scarciglia, A

    L. Scarciglia, A. Paolillo, and D. Palossi. A map-free deep learning-based framework for gate-to-gate monocular visual navigation aboard miniaturized aerial vehicles. In2025 IEEE International Conference on Robotics and Automation (ICRA), 2025. URLhttps://arxiv. org/abs/2503.05251

    work page arXiv 2025

  6. [6]

    Y . Qian, Y . Zhu, J. Luo, L. Liu, Y . Yuan, G. Ning, and H. Liao. Breaking the latency barrier: Synergistic perception and control for high-frequency 3D ultra- sound servoing. In2026 IEEE International Conference on Robotics and Automa- tion (ICRA), 2026. URLhttps://ras.papercept.net/conferences/conferences/ ICRA26/program/ICRA26_ContentListWeb_4.html

    2026

  7. [7]

    Z. Duan, Y . Zhang, S. Geng, G. Liu, J. Boedecker, and C. X. Lu. Fast ecot: Efficient embodied chain-of-thought via thoughts reuse. In2026 IEEE International Conference on Robotics and Automation (ICRA), 2026. URLhttps://arxiv.org/abs/2506.07639

    work page arXiv 2026

  8. [8]

    C. Chi, Z. Xu, S. Feng, E. Cousineau, Y . Du, B. Burchfiel, R. Tedrake, and S. Song. Dif- fusion policy: Visuomotor policy learning via action diffusion.The International Journal of Robotics Research, 2024. doi:10.1177/02783649241273668. URLhttps://doi.org/10. 1177/02783649241273668

    work page doi:10.1177/02783649241273668 2024

Show all 61 references
  1. [9]

    Z. Wang, Z. Li, A. Mandlekar, Z. Xu, J. Fan, Y . Narang, L. Fan, Y . Zhu, Y . Balaji, M. Zhou, M.-Y . Liu, and Y . Zeng. One-step diffusion policy: Fast visuomotor policies via diffusion distillation. InInternational Conference on Machine Learning, 2025. URL https://openreview...

    2025

  2. [10]

    Zhong, J

    Z. Zhong, J. Li, J. He, H. Yan, X. Gong, G. Zhao, Y . Cai, J. Gao, X. Yan, B. Liu, Y . Chen, L. Yang, and H. Li. Dualcot-vla: Visual-linguistic chain of thought via parallel reasoning for vision-language-action models.arXiv preprint arXiv:2603.22280, 2026. URLhttps: //arxiv.or...

    2026

  3. [11]

    Y . Duan, H. Yin, and D. Kragic. Real-time iteration scheme for diffusion policy. In2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2025. URL https://arxiv.org/abs/2508.05396. 9

    2025

  4. [12]

    Zhang, Y

    J. Zhang, Y . Guo, X. Chen, Y .-J. Wang, Y . Hu, C. Shi, and J. Chen. Hirt: Enhancing robotic control with hierarchical robot transformers. In P. Agrawal, O. Kroemer, and W. Burgard, editors,Proceedings of The 8th Conference on Robot Learning, volume 270 ofProceedings of Machi...

    2025

  5. [13]

    Q. Bu, H. Li, L. Chen, J. Cai, J. Zeng, H. Cui, M. Yao, and Y . Qiao. Towards synergistic, generalized and efficient dual-system for robotic manipulation, 2024. URLhttps://arxiv. org/abs/2410.08001

    2024

  6. [14]

    B. Han, J. Kim, and J. Jang. A dual process vla: Efficient robotic manipulation leveraging vlm. InConference on Robot Learning Workshop on Language and Robot Learning, pages 1–10,

  7. [15]

    URLhttps://arxiv.org/abs/2410.15549

  8. [16]

    J. Liu, P. Zhao, Z. Kong, X. Shen, P. Dong, F. Yang, L. Cui, H. Tang, G. Yuan, W. Niu, W. Zhang, X. Lin, G. Liu, Y . Wang, and D. Huang. When should a robot think? resource-aware reasoning via reinforcement learning for embodied robotic decision-making.arXiv preprint arXiv:260...

    2026 arXiv

  9. [17]

    Hu, J.-N

    Y . Hu, J.-N. Zaech, N. Nikolov, Y . Yao, S. Dey, G. Albanese, R. Detry, L. Van Gool, and D. Paudel. AR-VLA: True autoregressive action expert for vision-language-action models. In Robotics: Science and Systems, 2026. URLhttps://arxiv.org/abs/2603.10126

    2026 arXiv

  10. [18]

    J. Tang, Y . Sun, Y . Zhao, S. Yang, Y . Lin, Z. Zhang, J. Hou, Y . Lu, Z. Liu, and S. Han. VLASH: Real-time VLAs via future-state-aware asynchronous inference.arXiv preprint arXiv:2512.01031, 2025. URLhttps://arxiv.org/abs/2512.01031

    2025

  11. [19]

    Huang, Y

    Z. Huang, Y . Zhang, J. Liu, R. Song, C. Tang, and J. Ma. TIC-VLA: A think-in-control vision-language-action model for robot navigation in dynamic environments. InInternational Conference on Machine Learning, 2026. URLhttps://arxiv.org/abs/2602.02459

    2026 arXiv

  12. [21]

    URLhttps://arxiv.org/abs/2603.06480

    arXiv

  13. [22]

    H. Xu, Z. Liu, Y . Luomei, and F. Xu. Aerial vision-language navigation with a unified frame- work for spatial, temporal and embodied reasoning.arXiv preprint arXiv:2512.08639, 2025. URLhttps://arxiv.org/abs/2512.08639

    2025 arXiv

  14. [23]

    S. Wang, Y . Luo, X. Chen, A. Luo, D. Li, C. Liu, S. Chen, Y . Zhang, and J. Yu. VLingNav: Embodied navigation with adaptive reasoning and visual-assisted linguistic memory.arXiv preprint arXiv:2601.08665, 2026. URLhttps://arxiv.org/abs/2601.08665

    2026

  15. [24]

    Luo and X

    K. Luo and X. Ma. EmergeNav: Structured embodied inference for zero-shot vision-and- language navigation in continuous environments.arXiv preprint arXiv:2603.16947, 2026. URLhttps://arxiv.org/abs/2603.16947

    2026 arXiv

  16. [25]

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

    2025

  17. [26]

    GeForce RTX 4090 graphics cards for gaming.https://www.nvidia.com/ en-us/geforce/graphics-cards/40-series/rtx-4090/, 2026

    NVIDIA. GeForce RTX 4090 graphics cards for gaming.https://www.nvidia.com/ en-us/geforce/graphics-cards/40-series/rtx-4090/, 2026. Accessed: 2026-05-15. 10

    2026

  18. [27]

    Jetson AGX Orin for next-gen robotics.https://www.nvidia.com/en-us/ autonomous-machines/embedded-systems/jetson-orin/, 2026

    NVIDIA. Jetson AGX Orin for next-gen robotics.https://www.nvidia.com/en-us/ autonomous-machines/embedded-systems/jetson-orin/, 2026. Accessed: 2026-05- 15

    2026

  19. [28]

    D. Shah, B. Osinski, B. Ichter, and S. Levine. LM-nav: Robotic navigation with large pre- trained models of language, vision, and action. In6th Annual Conference on Robot Learning,

  20. [29]

    URLhttps://openreview.net/forum?id=UW5A3SweAH

  21. [30]

    Yokoyama, S

    N. Yokoyama, S. Qian, M. Chang, T.-Y . Yang, and W. Wang. Vlfm: Vision-language fron- tier maps for zero-shot semantic navigation. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 111–117. IEEE, 2024. doi:10.1109/ICRA57147.2024. 10610287. URLhttps:...

    2024 doi

  22. [31]

    Z. Yin, C. Cheng, , Y . Guo, and Z. Li. Navigation with vlm framework: Towards going to any language, 2025. URLhttps://arxiv.org/abs/2410.02787

    2025

  23. [32]

    Blukis, C

    V . Blukis, C. Paxton, D. Fox, A. Garg, and Y . Artzi. A persistent spatial semantic representation for high-level natural language instruction execution. InProceedings of The 5th Conference on Robot Learning, volume 164 ofProceedings of Machine Learning Research, pages 706–71...

    2022

  24. [33]

    de Heuvel, W

    J. de Heuvel, W. Shi, X. Zeng, and M. Bennewitz. Subgoal-driven navigation in dynamic en- vironments using attention-based deep reinforcement learning.Field Robotics, 4:182–210,

  25. [34]

    URLhttps://fieldrobotics.net/Field_Robotics/Volume_4_files/Vol4_ 07.pdf

  26. [35]

    W. Hu, Y . Zhou, and H. W. Ho. Hierarchical deep deterministic policy gradient for autonomous maze navigation of mobile robots.arXiv preprint arXiv:2508.04994, 2025. URLhttps: //arxiv.org/abs/2508.04994

    2025

  27. [36]

    C. H. Song, J. Wu, C. Washington, B. M. Sadler, W.-L. Chao, and Y . Su. LLM-Planner: Few-shot grounded planning for embodied agents with large language models. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 2998–3009, 2023. URL https://openacc...

    2023

  28. [37]

    M. Ahn, A. Brohan, N. Brown, Y . Chebotar, O. Cortes, B. David, C. Finn, C. Fu, K. Gopalakr- ishnan, K. Hausman, et al. Do as i can, not as i say: Grounding language in robotic affordances. InConference on Robot Learning (CoRL), 2022

    2022

  29. [38]

    Singh, V

    I. Singh, V . Blukis, A. Mousavian, A. Goyal, D. Xu, J. Tremblay, D. Fox, J. Thomason, and A. Garg. Progprompt: Generating situated robot task plans using large language models. In2023 IEEE International Conference on Robotics and Automation (ICRA), 2023. doi:10. 1109/ICRA4889...

    2023 doi

  30. [39]

    Dalal, T

    M. Dalal, T. Chiruvolu, D. S. Chaplot, and R. Salakhutdinov. Plan-seq-learn: Language model guided reinforcement learning for solving long horizon robotics tasks.arXiv preprint arXiv:2405.01534, 2024. URLhttps://arxiv.org/abs/2405.01534

    2024

  31. [40]

    M. F. Ginting, S.-K. Kim, D. D. Fan, M. Palieri, M. J. Kochenderfer, and A.-a. Agha- Mohammadi. Seek: Semantic reasoning for object goal navigation in real world inspection tasks. InRobotics: Science and Systems, 2024. URLhttps://arxiv.org/abs/2405. 09822

    2024

  32. [41]

    X. Zhou, T. Xiao, L. Liu, Y . Wang, M. Chen, X. Meng, X. Wang, W. Feng, W. Sui, and Z. Su. FSR-VLN: Fast and slow reasoning for vision-language navigation with hierarchical multi- modal scene graph.arXiv preprint arXiv:2509.13733, 2025. URLhttps://arxiv.org/ abs/2509.13733. 11

    2025

  33. [42]

    G. Li, N. Tsagkas, J. Song, R. Mon-Williams, S. Vijayakumar, K. Shao, and L. Sevilla- Lara. Learning precise affordances from egocentric videos for robotic manipulation. InProceedings of the IEEE/CVF International Conference on Computer Vision, 2025. URLhttps://openaccess.thec...

    2025

  34. [43]

    M. Pan, J. Zhang, T. Wu, Y . Zhao, W. Gao, and H. Dong. Omnimanip: Towards general robotic manipulation via object-centric interaction primitives as spatial constraints. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025. URLhttps://open...

    2025

  35. [44]

    Shridhar, J

    M. Shridhar, J. Thomason, D. Gordon, Y . Bisk, W. Han, R. Mottaghi, L. Zettlemoyer, and D. Fox. Alfred: A benchmark for interpreting grounded instructions for everyday tasks. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10...

    2020

  36. [45]

    C. Li, R. Zhang, J. Wong, C. Gokmen, S. Srivastava, R. Mart ´ın-Mart´ın, C. Wang, G. Levine, W. Ai, B. Martinez, H. Yin, M. Lingelbach, M. Hwang, A. Hiranaka, S. Garlanka, A. Ay- din, S. Lee, J. Sun, M. Anvari, M. Sharma, D. Bansal, S. Hunter, K.-Y . Kim, A. Lou, C. R. Matthew...

    2023

  37. [46]

    Brohan, N

    A. Brohan, N. Brown, J. Carbajal, Y . Chebotar, X. Chen, K. Choromanski, T. Ding, D. Driess, A. Fu, S. Guadarrama, et al. Rt-2: Vision-language-action models transfer knowledge to robotics.Conference on Robot Learning (CoRL), 2023

    2023

  38. [48]

    URLhttps://arxiv.org/abs/2410.24164

    arXiv

  39. [49]

    Belitsky, D

    M. Belitsky, D. J. Kopiczko, M. Dorkenwald, M. J. Mirza, J. R. Glass, C. G. M. Snoek, and Y . M. Asano. Kv cache steering for controlling frozen llms.arXiv preprint arXiv:2507.08799,

  40. [50]

    doi:10.48550/arXiv.2507.08799

    doi

  41. [51]

    Budzianowski, W

    P. Budzianowski, W. Maa, M. Freed, J. Mo, W. Hsiao, A. Xie, T. Mloduchowski, V . Tip- nis, and B. Bolte. Edgevla: Efficient vision-language-action models. InIROS 2024 Workshop on Mobile Manipulation and Embodied Intelligence, 2024. URLhttps:// mobile-manipulation.net/events/mo...

    2024

  42. [52]

    M. Xu, D. Niyato, H. Zhang, J. Kang, Z. Xiong, S. Mao, and Z. Han. Joint foundation model caching and inference of generative AI services for edge intelligence. In2023 IEEE Global Communications Conference (GLOBECOM), pages 3548–3553, 2023. doi:10.1109/ GLOBECOM54140.2023.1043...

    2023 doi

  43. [53]

    Huang, I

    H. Huang, I. Fedorov, A. Gromov, B. Beckerman, N. Suda, D. Eriksson, M. Balandat, R. Con- way, P. Huber, C. Sankar, A. Dalmia, Z. Liu, L. Wu, T. Elgamal, A. Sagar, V . Chandra, and R. Krishnamoorthi. MobileLLM-flash: Latency-guided on-device LLM design for industry scale. InPr...

    2026 arXiv

  44. [55]

    URLhttps://arxiv.org/abs/2510.14624

  45. [56]

    J. Li, M. Wu, J. Cao, A. Tiulpin, and M. B. Blaschko. EchoPrune: Interpreting redundancy as temporal echoes for efficient VideoLLMs.arXiv preprint arXiv:2605.10050, 2026. URL https://arxiv.org/abs/2605.10050

    2026 arXiv

  46. [57]

    H. Chen, J. Liu, C. Gu, Z. Liu, R. Zhang, X. Li, X. He, Y . Guo, C.-W. Fu, S. Zhang, and P.-A. Heng. Fast-in-slow: A dual-system vla model unifying fast manipulation within slow reasoning. InAdvances in Neural Information Processing Systems, 2025. URLhttps:// openreview.net/fo...

    2025

  47. [58]

    Sakurada, M

    K. Sakurada, M. Shibuya, and W. Wang. Weakly Supervised Silhouette-based Semantic Scene Change Detection. InIEEE International Conference on Robotics and Automation (ICRA), pages 6861–6867. IEEE, 2020. doi:10.1109/ICRA40945.2020.9196985

    2020 doi

  48. [59]

    B. Liu, Y . Liu, Z. Cui, S. Huang, and Y . Zhu. Libero: Benchmarking knowledge transfer in lifelong robot learning. InThirty-seventh Conference on Neural Information Processing Systems (NeurIPS), 2023

    2023

  49. [60]

    S. Bai, Y . Cai, R. Chen, K. Chen, X. Chen, Z. Cheng, L. Deng, W. Ding, C. Gao, C. Ge, et al. Qwen3-vl technical report.arXiv preprint arXiv:2511.21631, 2025. doi:10.48550/arXiv.2511. 21631. URLhttps://arxiv.org/abs/2511.21631

    2025 doi arXiv

  50. [61]

    Codepy-Deepthink-3B.https://huggingface.co/prithivMLmods/ Codepy-Deepthink-3B, 2024

    prithivMLmods. Codepy-Deepthink-3B.https://huggingface.co/prithivMLmods/ Codepy-Deepthink-3B, 2024. Hugging Face model card, accessed 2026-05-26

    2024

  51. [62]

    Llama-3.2-3B-Instruct.https://huggingface.co/meta-llama/Llama-3

    Meta. Llama-3.2-3B-Instruct.https://huggingface.co/meta-llama/Llama-3. 2-3B-Instruct, 2024. Hugging Face model card, accessed 2026-05-26

    2024

  52. [63]

    T. Wolf, L. Debut, V . Sanh, J. Chaumond, C. Delangue, A. Moi, P. Cistac, T. Rault, R. Louf, M. Funtowicz, J. Davison, S. Shleifer, P. von Platen, C. Ma, Y . Jernite, J. Plu, C. Xu, T. Le Scao, S. Gugger, M. Drame, Q. Lhoest, and A. Rush. Transformers: State-of-the-art natural...

    2020 doi

  53. [64]

    Instruction

    S.-A. Yu, F. Gao, Y . Wu, C. Yu, and Y . Wang. D3P: Dynamic denoising diffusion policy via reinforcement learning.arXiv preprint arXiv:2508.06804, 2025. URLhttps://arxiv.org/ abs/2508.06804. 13 A Appendix A.1 Affordance Router Details: Prompts and Conditions This section provi...

    2025

Pith tools

Reviewed June 28, 2026 · model on record in the stance chip above.