pith. sign in

arxiv: 2607.01088 · v1 · pith:XAYWHEAWnew · submitted 2026-07-01 · 💻 cs.RO · cs.DC

ROSA: A Robotics Foundation Model Serving System for Robot Factories

Wenqi Jiang , Jason Clemons , Rowland O'Flaherty , Hugo Hadfield , Alperen Degirmenci , Shuran Song , Yashraj Narang , Christos Kozyrakis This is my paper

Pith reviewed 2026-07-02 11:09 UTC · model grok-4.3

classification 💻 cs.RO cs.DC
keywords robotics foundation modelsmodel serving systemshared GPU poolfactory productivitymulti-robot fleetsscheduling for SLOsrobotics-aware abstractions
0
0 comments X

The pith

ROSA lets fleets of factory robots share server-class GPUs over the network to run robotics foundation models and raises overall productivity up to 12 times versus dedicated per-robot hardware.

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

The paper sets out to show that a serving system built for many robots in one factory can outperform the usual approach of giving each robot its own dedicated GPU. Instead of treating each inference request in isolation, ROSA pools GPUs on servers, supplies programming tools that handle chains of models and failures, and schedules work to maximize the total number of factory tasks that finish on time. A reader would care because robotics foundation models are making general-purpose robots viable for real factories, yet the serving layer determines whether those robots can scale without massive hardware costs or power drains. The evaluation on physical robots and large simulated workloads reports gains reaching 12.06 times in measured factory output.

Core claim

ROSA adopts shared GPU-pool serving so a fleet of robots can access powerful server-class GPUs over the network, supplies a robotics-aware programming abstraction that supports multi-model pipelines, per-task performance targets, and failure handling, and applies factory-objective-driven scheduling that maximizes the number of SLO-qualified tasks completed rather than minimizing latency for any single request. Built on Ray Serve with vLLM, PyTorch, and JAX backends, the system is tested on real robots and synthetic large-scale workloads and delivers up to 12.06 times higher factory productivity than conventional dedicated serving systems.

What carries the argument

Factory-objective-driven scheduling that allocates shared GPU resources across robot fleets while respecting per-task service level objectives and robotics-aware multi-model pipelines.

If this is right

  • A fleet can share a smaller number of high-end GPUs instead of equipping every robot with its own, improving utilization and extending battery life on the robots.
  • Multi-model pipelines and failure recovery become first-class features that the serving layer manages without each robot handling them locally.
  • Scheduling decisions are made to increase the count of tasks that meet factory-wide performance targets rather than to shorten any individual inference time.

Where Pith is reading between the lines

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

  • Factories could lower capital costs by deploying fewer total GPUs while still supporting larger robot fleets, provided the network remains stable.
  • The same shared-pool and productivity-driven approach might apply to other multi-robot settings such as warehouses if similar network conditions hold.
  • Direct measurements of end-to-end factory output under varying network quality would be the clearest way to confirm whether the reported gains survive real industrial conditions.

Load-bearing premise

Network latency and reliability problems in a real factory setting will not cancel out the advantages of shared GPU access, and the added robotics-aware abstractions will not create new failure modes that reduce total output.

What would settle it

A physical factory deployment in which measured task throughput under ROSA falls below twice the throughput of a dedicated-GPU baseline once real network delays and packet loss are present.

Figures

Figures reproduced from arXiv: 2607.01088 by Alperen Degirmenci, Christos Kozyrakis, Hugo Hadfield, Jason Clemons, Rowland O'Flaherty, Shuran Song, Wenqi Jiang, Yashraj Narang.

Figure 1
Figure 1. Figure 1: Robots working on various tasks in a factory. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: ROSA system overview. availability without requiring frequent recharging cycles. (A3) Improved hardware utilization and cost efficiency. With requests arriving from multiple robots, a shared serving system enables inter-robot batching, thereby improving GPU utiliza￾tion compared to one-to-one deployments that serve at most one request at a time. Furthermore, centralized serving systems eliminate the need t… view at source ↗
Figure 3
Figure 3. Figure 3: ROSA provides a declarative programming abstraction specifies the servers, the robot fleet, and detailed task requirements. lower than that of the inspection task (line 60). As another example, the safety-checking model in the inspection task is invoked twice per second and has a P99 latency requirement of 500 ms (line 72). Safety and SLO violations handling. Safety violations can arise either because a sa… view at source ↗
Figure 4
Figure 4. Figure 4: An example multi-model RFM inference pipeline. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Adaptive frequency search over heterogeneous tasks. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Single-task end-to-end performance measured as SLO-qualified action throughput. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: ROSA versus dedicated serving baselines. serving baselines on the single-task workloads described in [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: System 1 latency and SLO qualification on P4. [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 12
Figure 12. Figure 12: Performance of different resource allocations on P4. [PITH_FULL_IMAGE:figures/full_fig_p012_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: System 1 performance under different batch sizes. [PITH_FULL_IMAGE:figures/full_fig_p012_13.png] view at source ↗
Figure 15
Figure 15. Figure 15: Real-robot execution trace: the action model drives a Franka Panda arm to place tools into a bucket. [PITH_FULL_IMAGE:figures/full_fig_p013_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: VLM judges for monitoring task completion (left) and detecting nearby humans for safe operation (right). [PITH_FULL_IMAGE:figures/full_fig_p013_16.png] view at source ↗
read the original abstract

Robotics foundation models (RFMs) are making general-purpose robots increasingly practical for factory deployments. While RFM serving systems are central to this vision, existing systems are largely shaped by a single-robot, single-model assumption: inference is treated as an edge-computing problem handled by an on-robot or dedicated nearby GPU, and the serving objective is to minimize the latency of a single action model. In this paper, we propose ROSA, an RFM serving system for robot factories designed around three key principles. First, ROSA adopts shared GPU-pool serving, allowing a fleet of robots to access powerful server-class GPUs over the network in order to improve inference performance, battery duration, and GPU utilization. Second, ROSA provides a robotics-aware programming abstraction and system design that supports multi-model pipelines, per-task performance requirements, and failure handling. Third, ROSA uses factory-objective-driven scheduling to maximize SLO-qualified factory productivity rather than minimizing individual request latency. We implement ROSA on top of Ray Serve for distributed orchestration, with vLLM, PyTorch, and JAX as model-serving backends, and evaluate it on both real robots and synthetic large-scale workloads. The results show that ROSA improves factory productivity by up to 12.06x over conventional dedicated serving systems.

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 / 1 minor

Summary. The paper introduces ROSA, an RFM serving system for robot factories that replaces the single-robot dedicated-GPU model with three elements: shared GPU-pool serving over the network, robotics-aware abstractions supporting multi-model pipelines and failure handling, and factory-objective scheduling that maximizes SLO-qualified productivity rather than per-request latency. The system is built on Ray Serve with vLLM/PyTorch/JAX backends and is evaluated on real robots plus synthetic workloads, claiming up to 12.06x productivity gains over conventional dedicated serving systems.

Significance. If the empirical claims hold under realistic conditions, ROSA could materially improve GPU utilization, robot battery life, and factory throughput when deploying foundation models at scale. The shift from latency minimization to factory-level objective optimization is a substantive contribution to robotics systems research.

major comments (2)
  1. [Abstract] Abstract: the central 12.06x productivity claim is stated without any description of the workload, baseline implementation, measurement methodology, or error bars, rendering the result impossible to assess from the provided text.
  2. [Evaluation] Evaluation section: no end-to-end measurements are reported under realistic factory network conditions (variable latency, jitter, or transient disconnects), even though the architecture relies on network-based shared GPU access; if these factors increase missed SLOs or recovery overhead, the reported multiplier cannot be sustained.
minor comments (1)
  1. The terms 'conventional dedicated serving systems' and 'SLO-qualified factory productivity' are used without explicit definitions or references to prior work.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback. We address each major comment point-by-point below, indicating where revisions will be made to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central 12.06x productivity claim is stated without any description of the workload, baseline implementation, measurement methodology, or error bars, rendering the result impossible to assess from the provided text.

    Authors: We agree that the abstract's brevity omits these specifics, which are instead detailed in the Evaluation section (real-robot and synthetic workloads, dedicated per-robot baseline, productivity metric under SLOs, and reported gains). To address the concern, we will revise the abstract to include a concise clause referencing the evaluation setup and directing readers to the full methodology, while preserving the abstract's length constraints. revision: yes

  2. Referee: [Evaluation] Evaluation section: no end-to-end measurements are reported under realistic factory network conditions (variable latency, jitter, or transient disconnects), even though the architecture relies on network-based shared GPU access; if these factors increase missed SLOs or recovery overhead, the reported multiplier cannot be sustained.

    Authors: The real-robot experiments use network-based GPU access and thus incorporate some natural network variability, with the robotics-aware failure handling designed to manage disconnects. However, we did not perform controlled sweeps of latency/jitter or explicit transient disconnect scenarios. We will add a dedicated paragraph in the Evaluation section analyzing observed network effects from the real-robot runs and discussing implications for the productivity multiplier; if space and resources permit, we will also include targeted synthetic experiments quantifying sensitivity to these factors. revision: partial

Circularity Check

0 steps flagged

No significant circularity; empirical results are self-contained

full rationale

The paper reports an empirical performance claim (up to 12.06x factory productivity improvement) obtained from direct evaluations on real robots plus synthetic workloads using Ray Serve, vLLM, PyTorch, and JAX. No equations, fitted parameters, or first-principles derivations appear in the abstract or description. The productivity multiplier is presented as a measured outcome rather than a prediction derived from inputs by construction, and no load-bearing self-citations or self-definitional steps are indicated. The derivation chain therefore rests on external experimental data and does not reduce to its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no identifiable free parameters, axioms, or invented entities; the productivity claim rests on unspecified empirical evaluation.

pith-pipeline@v0.9.1-grok · 5790 in / 1061 out tokens · 39831 ms · 2026-07-02T11:09:30.596616+00:00 · methodology

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

52 extracted references · 33 canonical work pages · 20 internal anchors

  1. [1]

    Cosmos 3: Omnimodal World Models for Physical AI

    Niket Agarwal, Arslan Ali, Jon Allen, Martin Antolini, Adeline Aubame, Alisson Azzolini, Junjie Bai, Maciej Bala, Yogesh Balaji, Josh Bapst, et al. Cosmos 3: Omnimodal world models for physical ai.arXiv preprint arXiv:2606.02800, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  2. [2]

    DeepFleet: Multi-Agent Foundation Models for Mobile Robots

    Ameya Agaskar, Sriram Siva, William Pickering, Kyle O’Brien, Charles Kekeh, Alexandre Ormiga Galvao Bar- bosa, Ang Li, Brianna Gallo Sarker, Alicia Chua, Mayur Nemade, et al. Deepfleet: Multi-agent foundation models for mobile robots.arXiv preprint arXiv:2508.08574, 2025. 13

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  3. [3]

    Understanding Asynchronous Inference Methods for Vision-Language-Action Models

    Ayoub Agouzoul. Understanding asynchronous inference methods for vision-language-action models. arXiv preprint arXiv:2605.08168, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  4. [4]

    Performance of 802.11 be wi-fi 7 with multi-link operation on ar applications

    Molham Alsakati, Charlie Pettersson, Sebastian Max, Vishnu Narayanan Moothedath, and James Gross. Performance of 802.11 be wi-fi 7 with multi-link operation on ar applications. In2023 IEEE Wireless Communications and Networking Conference (WCNC), pages 1–6. IEEE, 2023

    2023

  5. [5]

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

    Ali Amin, Raichelle Aniceto, Ashwin Balakrishna, Kevin Black, Ken Conley, Grace Connors, James Darpinian, Karan Dhabalia, Jared DiCarlo, Danny Driess, et al. π0.6: a vla that learns from experience. arXiv preprint arXiv:2511.14759, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  6. [6]

    GR00T N1: An Open Foundation Model for Generalist Humanoid Robots

    Johan Bjorck, Fernando Castañeda, 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

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  7. [7]

    Johan Bjorck, Zhiqi Li, Yunze Man, Jing Wang, An-Chieh Cheng, Sifei Liu, Shihao Wang, Zhiding Yu, Abhishek Badki, Stan Birchfield, Valts Blukis, Yevgen Chebotar, Siyi Chen, Sicong Leng, Yu-Cheng Chou, Tianli Ding, Boyi Li, Zhengyi Luo, Hang Su, Jonathan Tremblay, Tingwu Wang, Bowen Wen, Jimmy Wu, Xianghui Xie, Hanrong Ye, Hongxu Yin, K. R. Zentner, Liangy...

    2026

  8. [8]

    $\pi_0$: 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, et al. π0: A visionlanguage-action flow model for general robot control, 2024a.URL https://arxiv. org/abs/2410.24164, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  9. [9]

    Real-Time Execution of Action Chunking Flow Policies

    Kevin Black, Manuel Y Galliker, and Sergey Levine. Real-time execution of action chunking flow policies. arXiv preprint arXiv:2506.07339, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  10. [10]

    Orca: An open-source, reliable, cost-effective, anthropomorphic robotic hand for uninterrupted dexterous task learning

    Clemens C Christoph, Maximilian Eberlein, Filippos Katsimalis, Arturo Roberti, Aristotelis Sympetheros, Michel R V ogt, Davide Liconti, Chenyu Yang, Barn- abas Gavin Cangan, Ronan J Hinchet, et al. Orca: An open-source, reliable, cost-effective, anthropomorphic robotic hand for uninterrupted dexterous task learning. arXiv preprint arXiv:2504.04259, 2025

    work page arXiv 2025

  11. [11]

    Kairos: A Scalable Serving System for Physical AI

    Yinwei Dai, Ganesh Ananthanarayanan, Landon Cox, Xenofon Foukas, Bozidar Radunovic, and Ravi Netravali. Kairos: A scalable serving system for physical ai.arXiv preprint arXiv:2605.11381, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  12. [12]

    Aha: A vision-language-model for detecting and reasoning over failures in robotic manipulation.arXiv preprint arXiv:2410.00371, 2024

    Jiafei Duan, Wilbert Pumacay, Nishanth Kumar, Yi Ru Wang, Shulin Tian, Wentao Yuan, Ranjay Krishna, Dieter Fox, Ajay Mandlekar, and Yijie Guo. Aha: A vision-language-model for detecting and reasoning over failures in robotic manipulation.arXiv preprint arXiv:2410.00371, 2024

    work page arXiv 2024

  13. [13]

    Score the steps, not just the goal: Vlm-based subgoal evaluation for robotic manipulation.arXiv preprint arXiv:2509.19524, 2025

    Ramy ElMallah, Krish Chhajer, and Chi-Guhn Lee. Score the steps, not just the goal: Vlm-based subgoal evaluation for robotic manipulation.arXiv preprint arXiv:2509.19524, 2025

    work page arXiv 2025

  14. [14]

    F.02 contributed to the production of 30,000 cars at bmw.https://www.figure.ai/news/ production-at-bmw, November 2025

    Figure AI. F.02 contributed to the production of 30,000 cars at bmw.https://www.figure.ai/news/ production-at-bmw, November 2025

    2025

  15. [15]

    Helix: A vision-language-action model for generalist humanoid control

    Figure AI. Helix: A vision-language-action model for generalist humanoid control. https: //www.figure.ai/news/helix, 2025

    2025

  16. [16]

    Fast-thinkact: Efficient vision-language-action reasoning via verbalizable latent planning.arXiv preprint arXiv:2601.09708, 2026

    Chi-Pin Huang, Yunze Man, Zhiding Yu, Min-Hung Chen, Jan Kautz, Yu-Chiang Frank Wang, and Fu-En Yang. Fast-thinkact: Efficient vision-language-action reasoning via verbalizable latent planning.arXiv preprint arXiv:2601.09708, 2026

    work page arXiv 2026

  17. [17]

    Enabling the robotic revolution: Bridging the performance gap between present and future

    Qijing Huang, Wenqi Jiang, Christos Kozyrakis, and Jason Clemons. Enabling the robotic revolution: Bridging the performance gap between present and future. In2026 IEEE/JSAP Symposium on VLSI Technology and Circuits (VLSI Technology and Circuits), Honolulu, HI, USA, 2026

    2026

  18. [18]

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

    P Intelligence, K Black, N Brown, J Darpinian, K Dha- balia, D Driess, A Esmail, M Equi, C Finn, N Fusai, et al. π0.5: A vision-language-action model with open-world generalization. arxiv 2025.arXiv preprint arXiv:2504.16054

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  19. [19]

    How fast can i run my vla? demystifying vla inference performance with vla-perf

    Wenqi Jiang, Jason Clemons, Karu Sankaralingam, and Christos Kozyrakis. How fast can i run my vla? demystifying vla inference performance with vla-perf. arXiv preprint arXiv:2602.18397, 2026

    work page arXiv 2026

  20. [20]

    Safety aware task planning via large language models in robotics

    Azal Ahmad Khan, Michael Andrev, Muhammad Ali Murtaza, Sergio Aguilera, Rui Zhang, Jie Ding, Seth Hutchinson, and Ali Anwar. Safety aware task planning via large language models in robotics. In2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 21024–21031. IEEE, 2025

    2025

  21. [21]

    Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success

    Moo Jin Kim, Chelsea Finn, and Percy Liang. Fine- tuning vision-language-action models: Optimizing speed and success.arXiv preprint arXiv:2502.19645, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  22. [22]

    OpenVLA: An Open-Source Vision-Language-Action Model

    Moo Jin Kim, Karl Pertsch, Siddharth Karamcheti, Ted Xiao, Ashwin Balakrishna, Suraj Nair, Rafael Rafailov, Ethan Foster, Grace Lam, Pannag Sanketi, et al. Openvla: 14 An open-source vision-language-action model.arXiv preprint arXiv:2406.09246, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  23. [23]

    Efficient memory management for large language model serving with pagedattention

    Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph Gonzalez, Hao Zhang, and Ion Stoica. Efficient memory management for large language model serving with pagedattention. In Proceedings of the 29th symposium on operating systems principles, pages 611–626, 2023

    2023

  24. [24]

    Amo: Adaptive motion optimization for hyper-dexterous humanoid whole- body control,

    Jialong Li, Xuxin Cheng, Tianshu Huang, Shiqi Yang, Ri- Zhao Qiu, and Xiaolong Wang. Amo: Adaptive motion optimization for hyper-dexterous humanoid whole-body control.arXiv preprint arXiv:2505.03738, 2025

    work page arXiv 2025

  25. [25]

    Evo-1: Lightweight vision-language-action model with preserved semantic alignment.arXiv preprint arXiv:2511.04555, 2025

    Tao Lin, Yilei Zhong, Yuxin Du, Jingjing Zhang, Jiting Liu, Yinxinyu Chen, Encheng Gu, Ziyan Liu, Hongyi Cai, Yanwen Zou, et al. Evo-1: Lightweight vision-language-action model with preserved semantic alignment.arXiv preprint arXiv:2511.04555, 2025

    work page arXiv 2025

  26. [26]

    When Should a Robot Think? Resource-Aware Reasoning via Reinforcement Learning for Embodied Robotic Decision-Making

    Jun Liu, Pu Zhao, Zhenglun Kong, Xuan Shen, Peiyan Dong, Fan Yang, Lin Cui, Hao Tang, Geng Yuan, Wei Niu, et al. When should a robot think? resource-aware rea- soning via reinforcement learning for embodied robotic decision-making.arXiv preprint arXiv:2603.16673, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  27. [27]

    A first look at wi-fi 6 in action: Throughput, latency, energy efficiency, and security.Proceedings of the ACM on Measurement and Analysis of Computing Systems, 7(1):1–25, 2023

    Ruofeng Liu and Nakjung Choi. A first look at wi-fi 6 in action: Throughput, latency, energy efficiency, and security.Proceedings of the ACM on Measurement and Analysis of Computing Systems, 7(1):1–25, 2023

    2023

  28. [28]

    Vision-language models for robot success detection

    Fiona Luo. Vision-language models for robot success detection. InProceedings of the AAAI Conference on Arti- ficial Intelligence, volume 38, pages 23750–23752, 2024

    2024

  29. [29]

    SONIC: Supersizing Motion Tracking for Natural Humanoid Whole-Body Control

    Zhengyi Luo, Ye Yuan, Tingwu Wang, Chenran Li, Sirui Chen, Fernando Castaneda, Zi-Ang Cao, Jiefeng Li, David Minor, Qingwei Ben, et al. Sonic: Supersizing motion tracking for natural humanoid whole-body control.arXiv preprint arXiv:2511.07820, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  30. [30]

    Running vlas at real-time speed.arXiv preprint arXiv:2510.26742, 2025

    Yunchao Ma, Yizhuang Zhou, Yunhuan Yang, Tiancai Wang, and Haoqiang Fan. Running vlas at real-time speed.arXiv preprint arXiv:2510.26742, 2025

    work page arXiv 2025

  31. [31]

    Ray: A distributed framework for emerging{AI} applications

    Philipp Moritz, Robert Nishihara, Stephanie Wang, Alexey Tumanov, Richard Liaw, Eric Liang, Melih Elibol, Zongheng Yang, William Paul, Michael I Jordan, et al. Ray: A distributed framework for emerging{AI} applications. In13th USENIX symposium on operating systems design and implementation (OSDI 18), pages 561–577, 2018

    2018

  32. [32]

    Chemist eye: a visual language model-powered system for safety monitoring and robot decision-making in self-driving laboratories.Digital Discovery, 5(5):2209–2220, 2026

    Francisco Munguia-Galeano, Zhengxue Zhou, Satheeshkumar Veeramani, Hatem Fakhruldeen, Louis Longley, Rob Clowes, and Andrew I Cooper. Chemist eye: a visual language model-powered system for safety monitoring and robot decision-making in self-driving laboratories.Digital Discovery, 5(5):2209–2220, 2026

    2026

  33. [33]

    Vulcan pick: A robotic system for picking targeted objects from fabric pods

    Kiru Park, Johannes Kulick, Alexander Melkozerov, Roc Arandes Vilagrasa, Teguh Santoso Lembono, Vanessa Neubauer, Artem Minichev, Kade Turner, Oana Agrigoroaiei, Pascal Klink, et al. Vulcan pick: A robotic system for picking targeted objects from fabric pods. 2025

    2025

  34. [34]

    Leave no observation behind: Real-time correction for vla action chunks.arXiv preprint arXiv:2509.23224, 2025

    Kohei Sendai, Maxime Alvarez, Tatsuya Matsushima, Yutaka Matsuo, and Yusuke Iwasawa. Leave no observation behind: Real-time correction for vla action chunks.arXiv preprint arXiv:2509.23224, 2025

    work page arXiv 2025

  35. [35]

    Lm-nav: Robotic navigation with large pre-trained models of language, vision, and action

    Dhruv Shah, Bła ˙zej Osi ´nski, Sergey Levine, et al. Lm-nav: Robotic navigation with large pre-trained models of language, vision, and action. InConference on robot learning, pages 492–504. pmlr, 2023

    2023

  36. [36]

    SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics

    Mustafa Shukor, Dana Aubakirova, Francesco Capuano, Pepijn Kooijmans, Steven Palma, Adil Zouitine, Michel Aractingi, Caroline Pascal, Martino Russi, Andres Marafioti, et al. Smolvla: A vision-language-action model for affordable and efficient robotics.arXiv preprint arXiv:2506.01844, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  37. [37]

    Hume: Introducing system-2 thinking in visual-language-action model.arXiv preprint arXiv:2505.21432, 2025

    Haoming Song, Delin Qu, Yuanqi Yao, Qizhi Chen, Qi Lv, Yiwen Tang, Modi Shi, Guanghui Ren, Maoqing Yao, Bin Zhao, et al. Hume: Introducing system-2 thinking in visual-language-action model.arXiv preprint arXiv:2505.21432, 2025

    work page arXiv 2025

  38. [38]

    Dadu-e: Rethinking the role of large language model in robotic computing pipelines.Journal of Field Robotics, 2026

    Wenhao Sun, Sai Hou, Zixuan Wang, Bo Yu, Shaoshan Liu, Xu Yang, Shuai Liang, Yiming Gan, and Yinhe Han. Dadu-e: Rethinking the role of large language model in robotic computing pipelines.Journal of Field Robotics, 2026

    2026

  39. [39]

    Vlash: Real-time vlas via future-state-aware asynchronous inference.arXiv preprint arXiv:2512.01031, 2025

    Jiaming Tang, Yufei Sun, Yilong Zhao, Shang Yang, Yujun Lin, Zhuoyang Zhang, James Hou, Yao Lu, Zhijian Liu, and Song Han. Vlash: Real-time vlas via future-state-aware asynchronous inference.arXiv preprint arXiv:2512.01031, 2025

    work page arXiv 2025

  40. [40]

    Gemini Robotics: Bringing AI into the Physical World

    Gemini Robotics Team, Saminda Abeyruwan, Joshua Ainslie, Jean-Baptiste Alayrac, Montserrat Gonzalez Arenas, Travis Armstrong, Ashwin Balakrishna, Robert Baruch, Maria Bauza, Michiel Blokzijl, et al. Gemini robotics: Bringing ai into the physical world.arXiv preprint arXiv:2503.20020, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  41. [41]

    Tesla to kill off model s and x ve- hicles, convert fremont factory to build robots

    Aidin Vaziri. Tesla to kill off model s and x ve- hicles, convert fremont factory to build robots. https://www.sfchronicle.com/tech/article/ tesla-end-model-s-x-21320796.php , January 2026. 15

    2026

  42. [42]

    Bitvla: 1-bit vision-language-action models for robotics manipulation.arXiv preprint arXiv:2506.07530, 2025

    Hongyu Wang, Chuyan Xiong, Ruiping Wang, and Xilin Chen. Bitvla: 1-bit vision-language-action models for robotics manipulation.arXiv preprint arXiv:2506.07530, 2025

    work page arXiv 2025

  43. [43]

    Probing Collision Grounding in Vision-Language Models for Safe Human-Robot Collaboration

    Jun Wang, Xiaohao Xu, and Xiaonan Huang. Probing collision grounding in vision-language models for safe human-robot collaboration.arXiv preprint arXiv:2605.31196, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  44. [44]

    Alpamayo-R1: Bridging Reasoning and Action Prediction for Generalizable Autonomous Driving in the Long Tail

    Yan Wang, Wenjie Luo, Junjie Bai, Yulong Cao, Tong Che, Ke Chen, Yuxiao Chen, Jenna Diamond, Yifan Ding, Wenhao Ding, et al. Alpamayo-r1: Bridging reasoning and action prediction for generalizable autonomous driving in the long tail.arXiv preprint arXiv:2511.00088, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  45. [45]

    Tinyvla: Towards fast, data-efficient vision-language-action models for robotic manipulation

    Junjie Wen, Yichen Zhu, Jinming Li, Minjie Zhu, Zhibin Tang, Kun Wu, Zhiyuan Xu, Ning Liu, Ran Cheng, Chaomin Shen, et al. Tinyvla: Towards fast, data-efficient vision-language-action models for robotic manipulation. IEEE Robotics and Automation Letters, 2025

    2025

  46. [46]

    Dysl-vla: Efficient vision-language-action model inference via dynamic-static layer-skipping for robot manipulation

    Zebin Yang, Yijiahao Qi, Tong Xie, Bo Yu, Shaoshan Liu, and Meng Li. Dysl-vla: Efficient vision-language-action model inference via dynamic-static layer-skipping for robot manipulation

  47. [47]

    World Action Models are Zero-shot Policies

    Seonghyeon Ye, Yunhao Ge, Kaiyuan Zheng, Shenyuan Gao, Sihyun Yu, George Kurian, Suneel Indupuru, You Liang Tan, Chuning Zhu, Jiannan Xiang, et al. World action models are zero-shot policies.arXiv preprint arXiv:2602.15922, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  48. [48]

    Deer- vla: Dynamic inference of multimodal large language models for efficient robot execution.Advances in Neural Information Processing Systems, 37:56619–56643, 2024

    Yang Yue, Yulin Wang, Bingyi Kang, Yizeng Han, Shen- zhi Wang, Shiji Song, Jiashi Feng, and Gao Huang. Deer- vla: Dynamic inference of multimodal large language models for efficient robot execution.Advances in Neural Information Processing Systems, 37:56619–56643, 2024

    2024

  49. [49]

    Hirt: Enhancing robotic control with hierarchical robot transformers.arXiv preprint arXiv:2410.05273, 2024

    Jianke Zhang, Yanjiang Guo, Xiaoyu Chen, Yen-Jen Wang, Yucheng Hu, Chengming Shi, and Jianyu Chen. Hirt: Enhancing robotic control with hierarchical robot transformers.arXiv preprint arXiv:2410.05273, 2024

    work page arXiv 2024

  50. [50]

    Qwen-RobotWorld Technical Report: Unifying Embodied World Modeling through Language-Conditioned Video Generation

    Jie Zhang, Xiaoyue Chen, Anzhe Chen, Chenxu Lv, Deqing Li, Gengze Zhou, Hang Yin, Haoqi Yuan, Haoyang Li, Jiahao Li, et al. Qwen-robotworld technical report: Unifying embodied world modeling through language-conditioned video generation.arXiv preprint arXiv:2606.17030, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  51. [51]

    Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware

    Tony Z Zhao, Vikash Kumar, Sergey Levine, and Chelsea Finn. Learning fine-grained bimanual manipulation with low-cost hardware.arXiv preprint arXiv:2304.13705, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  52. [52]

    Rt-2: Vision-language-action mod- els 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 mod- els transfer web knowledge to robotic control. InConfer- ence on Robot Learning, pages 2165–2183. PMLR, 2023. 16

    2023