Pith. sign in

REVIEW 4 major objections 6 minor 61 references

G0.5: One Autoregressive Stream for Robot Reasoning and Action

T0 review · 4 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A single autoregressive VLM that emits reasoning and action tokens in one stream can match or surpass VLM-as-encoder robot policies on seven benchmarks.

desk verdict Clear empirical case for single-stream autoregressive VLA, but the headline DROID number hides a marker-boosted task and the codec is unevaluated; deserves careful peer review. read the letter →

arxiv 2608.11739 v1 pith:CLKEL5D2 submitted 2026-08-12 cs.RO cs.AI

Yicheng Liu , Zibin Dong , Baijun Ye , Tianyuan Yuan , Tao Jiang , Anqi Yang , Shicheng Cao , Haonan Liu
show 19 more authors
Yue Sun Zihan Guo Xiao Liu Dong Ke Changxun Pan Chenru Wu Tailai Cheng Xiaoshu Ren Xinlei Zhang Jianning Cui Zijie Zhao Haoyu Zhang Kaiming Xu Haodong Yang Bowen Zhang Jiahui Niu Shaoting Zhu Shiduo Zhang Hang Zhao
This is my paper · ORCID
classification cs.ROcs.AI
keywords vision-language-actionmodelsautoregressivecontrolactiontokenizationcross-embodimentlearningchain-of-thoughtreasoningvisualmemoryrobotmanipulationreinforcementfine-tuning
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 argues that the dominant recipe for vision-language-action models—using a pretrained VLM only as a context encoder that feeds a separately trained flow-matching action expert—is unnecessary. It claims that a single autoregressive transformer producing reasoning tokens and action tokens in one stream, under one next-token objective, can match or surpass that recipe, and presents G0.5 as the concrete instance. Three components make this tractable: a learned cross-embodiment action codec that compresses continuous actions from many robots into shared discrete codes; a native chain-of-thought stream that interleaves subtask, bounding-box, trace, and action-hint tokens with action tokens under the same loss; and a visual-memory module that feeds multi-second history through the vision encoder. Across seven evaluation regimes, from real-robot fine-tuning to long-horizon household tasks in simulation, G0.5 matches or beats the strongest VLM-as-encoder baselines with a single set of weights, so if the paper is right the separate action expert is not needed for strong performance and the VLM's pretrained language abilities directly steer physical behavior.

What carries the argument

The load-bearing object is the token sequence template: a conditioning segment (multi-view RGB tokens, an embodiment identifier, the task instruction, proprioception, and a chain-of-thought prefix prompt, all without loss) followed by a generative segment on which the next-token cross-entropy loss is applied and which contains the optional chain-of-thought span and the action span. The action span unfolds as $R$ residual rounds, each emitting the currently active part markers (left control, right control, optionally lower body) followed by 8 action codes per part, which the cross-embodiment ActionCodec—a residual vector quantizer trained with a temporal contrastive objective over part-aligned action groups—decodes into continuous commands in a unified 27-dimensional space. This structure does three jobs at once: it compresses continuous actions so per-timestep autoregressive generation stays cheap, it shares one action vocabulary across embodiments so new robots require no new parameters, and it places reasoning tokens and action tokens in the same stream so reasoning can directly condition action prediction. Factorized spatial-temporal attention in the vision encoder supplies the multi-second visual memory that supports long-horizon closed-loop replanning.

What would settle it

Decode the ActionCodec's discrete tokens back into continuous commands on held-out DROID and R1-Lite trajectories and measure per-round end-effector and joint reconstruction error. If the error at the full $R$-round setting already exceeds grasp-level tolerance—on the order of a centimeter—then the token bottleneck, not the VLM, caps closed-loop performance, and the paper would need codec or residual-round scaling to close the gap; the paper currently reports no such numbers.

Watch

Extended reading notes

Core claim

The paper claims that a single autoregressive transformer, initialized from a pretrained vision-language model and trained with one next-token cross-entropy loss over a shared vocabulary, can generate both reasoning and actions in a single token stream, and in doing so match or surpass the VLM-as-encoder recipe in which a separate flow-matching expert produces the action distribution. The claim is carried by three mechanisms: the cross-embodiment ActionCodec, which maps continuous actions from 14 embodiments into a unified 27-dimensional part layout and emits 8 discrete codes per active part per residual round while dropping inactive parts from the stream; a native chain-of-thought span that supervises subtask text, object bounding boxes, 2D gripper traces, and action hints in the same loss as the action tokens; and factorized spatial-temporal attention that gives the vision encoder multi-second visual memory. The evidence spans seven regimes: 76.7% average success on R1-Lite/R1-Pro real-robot fine-tuning versus 53.3% for $\pi_{0.5}$ and 24.4% for GR00T-N1.7; a single-checkpoint Task Success Score of 0.3136 on the 50-task 2025 BEHAVIOR Challenge versus 0.2626 for $\pi_{0.5}$ at four epochs and 0.2605 for the four-checkpoint winner; 82.5% on DROID post-training with the evaluation environment and object instances held out; 98.9% on LIBERO; 93.3% on RoboTwin 2.0; 87.3% on SimplerEnv-Bridge; and a new Pick-and-Place benchmark where G0.5 leads at every post-training scale. The paper further claims that autoregressive action tokens expose exact log-probabilities, so GRPO-style reinforcement learning applies without reformulation and converges faster and more stably than the flow-matching variant, and that per-stage prompt wording visibly shifts zero-shot rollouts without retraining.

Load-bearing premise

The load-bearing premise is that the learned action tokenizer rebuilds continuous robot commands accurately enough for closed-loop control: the model emits only 8 discrete codes per active part per residual round, and if that compression loses contact-relevant detail, every benchmark result is capped by the tokenizer—yet the paper reports no reconstruction-error measurements and no ablations of residual-round count or codebook size.

Editorial extensions

If this is right

  • The separate flow-matching action expert used by $\pi_0$, $\pi_{0.5}$, GR00T, and SmolVLA is not required for strong VLA performance: a single autoregressive stream matches or exceeds those models on every one of the seven evaluation regimes reported.
  • Chain-of-thought stops being bolt-on: subtask, bounding-box, trace, and action-hint tokens are supervised in the same loss as actions, and turning CoT on improves grounding and execution on long-horizon, stage-conditioned tasks outside the pretraining distribution.
  • Autoregressive VLAs are natively RL-trainable: exact token-level log-probabilities let GRPO-style algorithms apply without reformulation, and in the paper's one-demonstration LIBERO probe the AR policy converges faster, higher, and with lower variance than the flow-matching variant.
  • Prompt wording becomes a control channel: adverbial qualifiers, spatial cues, and near-synonymous verbs visibly shift zero-shot rollouts without retraining, reported by the paper as a preliminary qualitative result of keeping the VLM's in-context capacity wired to action generation.
  • Single-checkpoint generalist policies suffice for long-horizon household tasks: one G0.5 checkpoint post-trained for a single epoch scores 0.2904 on the 50-task BEHAVIOR Challenge, above $\pi_{0.5}$ at four epochs (0.2626) and the four-checkpoint first-place solution (0.2605).

Reading between the lines

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

  • The paper's central comparison leaves the codec unmeasured, so a natural next experiment is to publish per-round reconstruction error on DROID and R1-Lite trajectories; if the codec already loses centimeter-level precision, then the ceiling on contact-rich tasks is the tokenizer, and scaling residual rounds or codebook size becomes the predicted lever.
  • The decoding-interface hypothesis—that autoregressive action tokens attend to the chain-of-thought directly while the flow-matching head sees a pooled summary—is testable on the released checkpoint by ablating attention masking between CoT and action tokens, which would separate reasoning quality from interface effects.
  • The qualitative prompt-steering results suggest a falsifiable program: if verb and adverb substitutions reliably shift action distributions on held-out scenes, action generation behaves like in-context learning, and systematic probing could turn prompt-level steering into a measurable capability.
  • Because the BEHAVIOR gap on container-interaction tasks tracks pretraining data composition rather than architecture, enriching the pretraining mixture with cabinet-and-appliance skills is a concrete extension the paper itself points to, and it predicts where the AR recipe's remaining weaknesses lie.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes G0.5, an autoregressive vision-language-action model initialized from Qwen3.5 2B, in which a single transformer decoder generates optional chain-of-thought tokens and discrete action codes under one next-token cross-entropy loss. Actions are encoded by a learned cross-embodiment residual-vector-quantization ActionCodec into R residual rounds of 8 codes per active DoF group, and future visual context is injected through a factorized visual-memory module. The model is pretrained on a mixture of robot demonstrations and VQA data, then evaluated across real-world fine-tuning on R1-Lite/R1-Pro, the BEHAVIOR Challenge, DROID post-training with zero-shot transfer, a new Pick-and-Place benchmark, and the LIBERO, RoboTwin 2.0, and SimplerEnv-Bridge simulation suites. The central claim is that a single autoregressive stream for reasoning and action can match or surpass VLM-as-encoder models with separate flow-matching action experts.

Significance. If the central claim holds, the paper provides a substantial data point in the VLA architecture debate: it would show that the flow-matching action expert is not necessary for strong performance and that autoregressive VLA training can preserve and exploit VLM reasoning. The evaluation breadth is a genuine strength: the manuscript uses multiple external benchmarks, reports controlled comparisons (e.g., the DROID marker study and the AR-vs-FM inference probe), and explicitly acknowledges open limitations such as drawer-insertion weakness, limited visual-memory horizon, and the qualitative nature of prompt-steering observations. The release of a pretrained backbone is also a practical contribution. However, two load-bearing gaps temper the significance: the paper never measures the fidelity of its learned ActionCodec or ablates its residual-round/codebook design, and the headline DROID result is obtained with added visual markers that change the comparison on the very tasks where the model is weakest. These gaps need to be closed before the architectural conclusion is fully supported.

major comments (4)
  1. [Section 3.1 and Figure 2] The entire policy output path is mediated by the learned ActionCodec: continuous action chunks are mapped to R residual rounds of 8 discrete codes per active DoF group, and the default policy has no other output route. The paper reports no action reconstruction error, no ablation over R or codebook size, and no measurement of how tokenization error propagates to task success. This is load-bearing because the benchmark comparisons against pi0.5/GR00T and the AR-vs-FM probe in Section 5.6 cannot, without these numbers, distinguish an advantage of the autoregressive interface from an accident of tokenizer capacity. The acknowledged drawer-insertion weakness in Section 6 could equally be a codec or a perception failure, and the manuscript gives no way to tell. Please add per-embodiment codec reconstruction metrics, an ablation of R and codebook size on at least one benchmark, and an analysis of how tokenization loss translates into closed-loop success.
  2. [Section 5.1.2 and Figure 6] The DROID average of 82.5% is computed on a setup in which orange adhesive markers were attached to the drawer for the towel-insertion task; the controlled comparison in Section 5.1.2 shows that G0.5 drops from 100% to 60% without the markers, while pi0.5-DROID stays at 90% and is described as 'relatively unaffected.' The claim that G0.5 'outperforms pi0.5 on all 10 tasks' and the 25.0-point margin therefore depend on a marker-assisted evaluation on a task class where the model is otherwise weaker than the baselines. Because this result appears in the abstract as a headline number, the manuscript should report DROID averages computed with and without markers and should label the main result as marker-assisted, or present the unassisted drawer tasks separately.
  3. [Section 5.6 and Figure 11] The AR-vs-FM inference probe uses only n=5 rollouts per cell for the two long-horizon tasks (Air Fryer and Cook Bacon), and the CoT traces in the AR and FM rollouts are not identical because the rollouts diverge once actions are executed. The claim that 'the AR head benefits more from CoT than the FM head' and the associated language-following gaps (72 vs. 48, 64 vs. 44) are therefore based on very small samples with uncontrolled divergence between conditions. The hand-scored CoT-correctness check is helpful, but it does not remove the confound. Please provide confidence intervals or significance information, increase the number of rollouts, and, if possible, run head-to-head comparisons on matched CoT traces to support the decoding-interface hypothesis.
  4. [Section 5.7 and Figure 12] The GRPO fine-tuning comparison is presented as evidence that the autoregressive interface is easier to optimize with RL, but the protocol selects four tasks on which AR and FM 'reached comparable initial success rates' after post-training with a single demonstration. This selection, combined with the approximate likelihood for the FM policy via an SDE reformulation following RLinf, may bias the comparison in favor of AR: the approximation noise and extra hyperparameters could explain the slower FM convergence. Please report the selection criterion and all tasks considered, show results for the full set, and provide a sensitivity analysis of the FM results to the RLinf approximation. Without this, 'RL-native AR > FM' is not established.
minor comments (6)
  1. [Abstract and Section 5.1.2] The abstract reports 82.5% on DROID without mentioning the orange adhesive markers that the controlled comparison shows are crucial for G0.5 on the drawer-insertion task; the caveat should be visible wherever the headline number is stated.
  2. [Section 4] The text says 'Evaluation uses the fixed no-CoT format,' which conflicts with Section 5.6, where CoT is toggled at inference. Please clarify that the fixed no-CoT format applies to the main benchmark evaluations and that the CoT toggle is a separate inference-time probe.
  3. [Section 5.6] The final sentence of the qualitative instruction-wording paragraph, 'This probe is AR-only,' is ambiguous after the AR/FM toggling described earlier in the same section; specify that the qualitative wording observations used only AR+CoT.
  4. [Figure 2] The chat-role wrappers <chat_user_prefix/suffix> and <chat_assistant_prefix/suffix> are not formally defined in the text; a one-sentence definition would help readers map the template to Eq. (1).
  5. [Section 5.3, Tables 4 and 6] The BEHAVIOR Task Success Score is reported as a mean over two evaluation runs, but no variance or per-run breakdown is given; adding this would strengthen the comparison with pi0.5, which is also a single-checkpoint result.
  6. [Section 5.5] The PP Bench description states that the evaluation uses 64 real-world trials, and the numbers in Figure 10 are consistent with that, but the text should state whether the 64 trials cover all 48 in-distribution and 16 out-of-distribution categories exactly once per model, to make the paired protocol fully transparent.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claim is an empirical benchmark comparison, and the authors' self-citations supply components and baselines rather than load-bearing derivations.

full rationale

The paper's central claim is empirical rather than derivational: G0.5's headline results are measured on external or held-out benchmarks (LIBERO, RoboTwin, SimplerEnv, DROID with a held-out environment and objects, BEHAVIOR) or on internally collected data with an explicit train/evaluation split. None of the reported success rates is produced by substituting the model's own fitted parameters back into the evaluation objective, and no equation equates the predicted outcome with the training loss or with the codec's reconstruction target. The FASTer and ActionCodec citations are component provenance: the learned tokenizer is an input to the system, not the quantity being predicted, and the external benchmark results would provide evidence against the method if the codec were inadequate. The AR-versus-FM probe is a controlled experiment comparing two decoders on the same checkpoint and on zero-shot long-horizon tasks, and the paper separately hand-scores CoT correctness to separate reasoning quality from the decoding interface, so the comparison is not defined into the result. The acknowledged lack of codec reconstruction error and residual-round ablations is an unverified assumption or reporting gap, not a circular reduction. No load-bearing step invokes an author-uniqueness theorem, and no fitted value is renamed as a prediction. The self-citations are real but not load-bearing: the central claim is independently supported by the external benchmark comparisons.

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

The paper is an empirical systems paper, so the free parameters are the hand-chosen design choices that define the action token budget and CoT supervision. The axioms are the unverified assumptions about codec fidelity, autolabel quality, and benchmark comparability that the results depend on. No new physical entities are postulated.

free parameters (3)
  • Unified action space partition (27-dim) = left_control 9, left_gripper 1, right_control 9, right_gripper 1, lower_body 7
    Chosen by hand to cover 14 embodiments; inactive slots are padded with noop tokens. The capacity of the action token stream is set by this partition, so it bounds what the codec can represent.
  • Residual rounds R and 8 action codes per group-round = R not stated; 8 codes per active DoF group per round
    Codec token budget chosen by hand. The paper does not report an ablation on R or codebook size and does not report reconstruction error, so the adequacy of this budget for downstream control is assumed.
  • Chain-of-thought sampling weights = Subtask format higher weight; other formats default
    Hand-chosen distribution over the 8 CoT formats during pretraining. It shapes what the model learns to generate before actions and is not ablated.
assumptions (3)
  • domain assumption The learned ActionCodec reconstructs continuous actions with sufficient fidelity for closed-loop control
    Invoked in Section 3.1: the VLM predicts discrete codes from this codec. If reconstruction error is large for contact-rich or high-frequency motions, the benchmark results would not hold. No reconstruction error metric or codec ablation is reported.
  • domain assumption The autolabeled CoT annotations are accurate enough to supervise reasoning
    Section 4: subtask, bounding box, trace, and action-hint labels are produced by an automated pipeline using Gemini 3, Doubao Seed 2.0, and SAM3. No quality control or human validation rate is reported; noisy labels could inflate or deflate the measured CoT benefits.
  • domain assumption Benchmark protocols measure what the paper claims
    The paper follows standard protocols for LIBERO, SimplerEnv, and RoboTwin, but for BEHAVIOR it follows the first-place solution's setup while omitting its stage head, and for SimplerEnv it compiles baseline numbers from other papers. Comparability is assumed across these choices.
invented entities (2)
  • Cross-embodiment ActionCodec independent evidence
    purpose: Learned RVQ action tokenizer with active-DoF grouping that maps heterogeneous continuous robot actions into a shared discrete token vocabulary.
    The codec has downstream evidence from the benchmark results and the AR-versus-FM comparisons, though no standalone reconstruction evaluation is provided in this paper.
  • Native chain-of-thought span independent evidence
    purpose: Intermediate reasoning tokens (Subtask, BBox, Trace, ActionHint) emitted in the same autoregressive stream as action tokens.
    Evaluated through CoT on/off comparisons and hand-scored correctness in Section 5.6, though the long-horizon probe uses only five rollouts per cell.

how reviews work

0 comments
Cite this review

Pith. "Pith review of G0.5: One Autoregressive Stream for Robot Reasoning and Action." pith.science (2026). https://pith.science/paper/CLKEL5D2

@misc{pith2026260811739,
  author       = {Pith},
  title        = {Pith review of: G0.5: One Autoregressive Stream for Robot Reasoning and Action},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CLKEL5D2}},
  note         = {Machine review of arXiv:2608.11739}
}
abstract

The prevailing recipe for Vision-Language-Action (VLA) models couples a pretrained VLM with a separately trained flow-matching action expert. This makes the VLM a context encoder rather than a decision-maker. We introduce G0.5, a pretrained autoregressive VLA in which a single transformer decoder emits reasoning and action tokens under a single objective. Three components make this tractable at foundation-model scale: a learnable cross-embodiment action tokenizer that maps heterogeneous robot actions into a shared vocabulary; a native chain-of-thought stream interleaving task decomposition, object grounding, and action hints with action tokens; and a visual memory module that injects multi-second history through the vision encoder. Because reasoning and action share a single set of weights, the pretrained VLM's capabilities carry over to physical behavior: the model follows instructions closely, and prompts directly steer action granularity, task horizon, and out-of-distribution scene handling without further training. Pretrained on a large collection of robot datasets together with VQA samples, G0.5 surpasses state-of-the-art models across 7 independent regimes: real-world fine-tuning on R1lite and R1pro robots (76.7\% vs.\ 53.3\% for $\pi_{0.5}$ and 24.4\% for GR00T-N1.7), the 2025 BEHAVIOR Challenge on 50 long-horizon household mobile manipulation tasks using a generalist policy (31.4\% vs.\ 26.3\% for $\pi_{0.5}$ and 26.1\% for the challenge winner), DROID post-training followed by zero-shot transfer to an unseen environment and objects (82.5\%), a language-following Pick-and-Place benchmark, LIBERO (98.9\%), RoboTwin 2.0 (93.3\%), and SimplerEnv-Bridge (87.3\%).

Figures

Figures reproduced from arXiv: 2608.11739 by the authors.

Figure 1
Figure 1. G0.5: reasoning and action in one autoregressive stream. Top left: A single VLM conditions on multi-view images, an instruction, proprioception, and an embodiment identifier, then generates optional chain-of-thought (CoT) and compact action codes under the same next-token objective. A cross-embodiment ActionCodec converts the codes into continuous motor commands, while active-part tokenization omits inactive control… view at source ↗
Figure 2
Figure 2. Token sequence template. All inputs and outputs are serialised into a single autoregressive sequence: a conditioning segment (multi-view RGB, embodiment id, task instruction, proprioceptive state—in user-side chat tokens) and a generative segment on which the next-token cross-entropy loss in Eq. (1) is applied. The generative segment composes an optional chain-of-thought span—any subset of four self-describing reaso… view at source ↗
Figure 3
Figure 3. Structured action tokenization. Heterogeneous robot actions are decomposed into semantically aligned motion parts, encoded with a residual vector quantizer, and serialized as part-specific action tokens. This representation shares one action vocabulary across embodiments while allowing sparse prediction over only the activated parts. 3.1 Structured Tokenization of Heterogeneous Action Data A key challenge is how to … view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Action and object concept distribution in the pre-training corpus. We show the most frequent action verbs and object nouns extracted from the pre-training data; each bar gives the concept’s share of occurrences (%) on a logarithmic axis, and the trailing Others bar agg…
Figure 5
Figure 5. Figure 5: DROID environment- and object-level zero-shot evaluation tasks. After DROID post-training, we evaluate G0.5 on 10 manipulation tasks across 8 held-out scene setups and previously unseen physical object instances on a Franka Research 3 robot arm. Tasks cover object plac…
Figure 6
Figure 6. Figure 6: DROID environment- and object-level zero-shot evaluation results. All models are trained on DROID data, while the physical evaluation environment and object instances are held out. Per-task success rates (%) are reported for 𝜋0.5-DROID, MolmoAct2-DROID, and G0.5 across…
Figure 7
Figure 7. Figure 7: Overview of real-world fine-tuning evaluation settings. We evaluate four manipulation tasks instantiated as six task-embodiment settings across R1-Pro and R1-Lite. The R1-Pro settings include towel folding, carton folding, and box transfer and stacking, while the R1-Li…
Figure 8
Figure 8. Figure 8: Performance on real-world fine-tuning evaluation. We evaluate G0.5 against 𝜋0.5 and GR00T-N1.7 on four manipulation tasks instantiated as six task-embodiment settings across R1-Pro and R1-Lite. G0.5 achieves strong overall performance in both task success rate and proc…
Figure 9
Figure 9. Figure 9: Pick-and-Place Benchmark setting. The evaluation set contains 64 object categories and 3 container categories. Each trial presents 16 randomly arranged objects and containers, requiring the robot to identify the instructed target among distractors and place it into the…
Figure 10
Figure 10. Figure 10: PP Bench evaluation results. Left: language following rate. Right: task success rate. We compare G0.5 and 𝜋0.5 across zero-shot, 1H, 10H, and 50H post-training settings. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_10.png]
Figure 11
Figure 11. Figure 11: Long-horizon zero-shot probe on Air Fryer and Cook Bacon (same pretrained G0.5 checkpoint with only inference-time switching). Progress score (0–5, number of completed sub-stages) and language-following rate across the four decoder × CoT cells: both metrics show the s…
Figure 12
Figure 12. Figure 12: GRPO fine-tuning: AR vs. FM. Average success rate over four LIBERO tasks during GRPO fine-tuning from a single demonstration per task (mean ± std over seeds). Starting from AR and FM policies with comparable initial success rates, the AR policy converges faster, attai…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

61 extracted references · 15 canonical work pages

  1. [1]

    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

    2023

  2. [2]

    Openvla: An open-source vision-language- action model.arXiv preprint arXiv:2406.09246, 2024

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

    arXiv 2024

  3. [3]

    Kevin Black, Noah Brown, Danny Driess, Adnan Esmail, Michael Equi, Chelsea Finn, Niccolo Fusai, Lachy Groom, Karol Hausman, Brian Ichter, et al.𝜋0: A vision-language-action flow model for general robot control.arXiv preprint arXiv:2410.24164, 2024

    arXiv

  4. [4]

    Physical Intelligence, Kevin Black, Noah Brown, James Darpinian, Karan Dhabalia, Danny Driess, Adnan Esmail, 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

    arXiv 2025

  5. [5]

    Gr00t n1: An open foundation model for generalist humanoid robots.arXiv preprint arXiv:2503.14734, 2025

    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

    arXiv 2025

  6. [6]

    Smolvla: A vision-language- action model for affordable and efficient robotics.arXiv preprint arXiv:2506.01844, 2025

    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

    arXiv 2025

  7. [7]

    Cot-vla: Visualchain-of-thoughtreasoningforvision-language-action models

    Qingqing Zhao, Yao Lu, Moo Jin Kim, Zipeng Fu, Zhuoyang Zhang, Yecheng Wu, Zhaoshuo Li, Qianli Ma,SongHan,ChelseaFinn,etal. Cot-vla: Visualchain-of-thoughtreasoningforvision-language-action models. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 1702–1713, 2025

    work page 2025

  8. [8]

    Dualcot-vla: Visual-linguistic chain of thought via parallel reasoning for vision-language-action models.arXiv preprint arXiv:2603.22280, 2026

    Zhide Zhong, Junfeng Li, Junjie He, Haodong Yan, Xin Gong, Guanyi Zhao, Yingjie Cai, Jiantao Gao, Xu Yan, Bingbing Liu, et al. Dualcot-vla: Visual-linguistic chain of thought via parallel reasoning for vision-language-action models.arXiv preprint arXiv:2603.22280, 2026

    arXiv 2026

Show all 61 references
  1. [9]

    Halo: A unified vision-language-action model for embodied multimodal chain-of-thought reasoning.arXiv preprint arXiv:2602.21157, 2026

    Quanxin Shou, Fangqi Zhu, Shawn Chen, Puxin Yan, Zhengyang Yan, Yikun Miao, Xiaoyi Pang, Zicong Hong, Ruikai Shi, Hao Huang, et al. Halo: A unified vision-language-action model for embodied multimodal chain-of-thought reasoning.arXiv preprint arXiv:2602.21157, 2026

    2026

  2. [10]

    Mem: Multi-scale embodied memory for vision language action models.arXiv preprint arXiv:2603.03596, 2026

    Marcel Torne, Karl Pertsch, Homer Walke, Kyle Vedder, Suraj Nair, Brian Ichter, Allen Z Ren, Haohuan Wang, Jiaming Tang, Kyle Stachowicz, et al. Mem: Multi-scale embodied memory for vision language action models.arXiv preprint arXiv:2603.03596, 2026

    2026

  3. [11]

    Fast: Efficient action tokenization for vision-language-action models.arXiv preprint arXiv:2501.09747, 2025

    Karl Pertsch, Kyle Stachowicz, Brian Ichter, Danny Driess, Suraj Nair, Quan Vuong, Oier Mees, Chelsea Finn, and Sergey Levine. Fast: Efficient action tokenization for vision-language-action models.arXiv preprint arXiv:2501.09747, 2025

    2025 arXiv

  4. [12]

    Flowvla: Visual chain of thought-based motion reasoning for vision-language-action models.arXiv preprint arXiv:2508.18269, 2025

    Zhide Zhong, Haodong Yan, Junfeng Li, Xiangchen Liu, Xin Gong, Tianran Zhang, Wenxuan Song, Jiayi Chen, Xinhu Zheng, Hesheng Wang, et al. Flowvla: Visual chain of thought-based motion reasoning for vision-language-action models.arXiv preprint arXiv:2508.18269, 2025

    2025

  5. [13]

    BEHAVIOR-1K: A benchmark for embodied ai with 1,000 everyday activities and realistic simulation

    Chengshu Li, Ruohan Zhang, Josiah Wong, Cem Gokmen, Sanjana Srivastava, Roberto Martín-Martín, Chen Wang, Gabrael Levine, Michael Lingelbach, Jiankai Sun, et al. BEHAVIOR-1K: A benchmark for embodied ai with 1,000 everyday activities and realistic simulation. InConference on R...

    2023

  6. [14]

    LIBERO: Benchmarking knowledge transfer for lifelong robot learning.Advances in Neural Information Processing Systems, 36:44776–44791, 2023

    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. 26

    2023

  7. [15]

    Xuanlin Li, Kyle Hsu, Jiayuan Gu, Karl Pertsch, Oier Mees, Homer Rich Walke, Chuyuan Fu, Ishikaa Lunawat, Isabel Sieh, Sean Kirmani, Sergey Levine, Jiajun Wu, Chelsea Finn, Hao Su, Quan Vuong, and TedXiao.Evaluatingreal-worldrobotmanipulationpoliciesinsimulation.arXivpreprinta...

    2024 arXiv

  8. [16]

    Fast-wam: Do world action models need test-time future imagination?arXiv preprint arXiv:2603.16666, 2026

    Tianyuan Yuan, Zibin Dong, Yicheng Liu, and Hang Zhao. Fast-wam: Do world action models need test-time future imagination?arXiv preprint arXiv:2603.16666, 2026

    2026 arXiv

  9. [17]

    Knowledge insulating vision-language-action models: Trainfast, runfast, generalizebetter.AdvancesinNeuralInformationProcessingSystems, 38:102867–102888, 2026

    Danny Driess, Jost Springenberg, Brian Ichter, Lili Yu, Adrian Li-Bell, Karl Pertsch, Allen Ren, Homer Walke, Quan Vuong, Lucy Xiaoyang Shi, et al. Knowledge insulating vision-language-action models: Trainfast, runfast, generalizebetter.AdvancesinNeuralInformationProcessingSys...

    2026

  10. [18]

    AnkitGoyal,HugoHadfield,XuningYang,ValtsBlukis,andFabioRamos.Vla-0: Buildingstate-of-the-art vlas with zero modification.arXiv preprint arXiv:2510.13054, 2025

    2025

  11. [19]

    Vq-vla: Improving vision-language-action models via scaling vector-quantized action tokenizers

    Yating Wang, Haoyi Zhu, Mingyu Liu, Jiange Yang, Hao-Shu Fang, and Tong He. Vq-vla: Improving vision-language-action models via scaling vector-quantized action tokenizers. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 11089–11099, 2025

    2025

  12. [20]

    Beast: Efficient tokenization of b-splines encoded action sequences for imitation learning.Advances in Neural Information Processing Systems, 38:172934–172959, 2026

    Hongyi Zhou, Weiran Liao, Xi Huang, Yucheng Tang, Fabian Otto, Xiaogang Jia, Xinkai Jiang, Simon Hilber, Ge Li, Qian Wang, et al. Beast: Efficient tokenization of b-splines encoded action sequences for imitation learning.Advances in Neural Information Processing Systems, 38:17...

    2026

  13. [21]

    Behavior generation with latent actions.arXiv preprint arXiv:2403.03181, 2024

    Seungjae Lee, Yibin Wang, Haritheja Etukuru, H Jin Kim, Nur Muhammad Mahi Shafiullah, and Lerrel Pinto. Behavior generation with latent actions.arXiv preprint arXiv:2403.03181, 2024

    2024 arXiv

  14. [22]

    Spatialvla: Exploring spatial representations for visual-language-action model

    Delin Qu, Haoming Song, Qizhi Chen, Yuanqi Yao, Xinyi Ye, Yan Ding, Zhigang Wang, JiaYuan Gu, Bin Zhao, Dong Wang, et al. Spatialvla: Exploring spatial representations for visual-language-action model. arXiv preprint arXiv:2501.15830, 2025

    2025 arXiv

  15. [23]

    Being-h0

    Hao Luo, Ye Wang, Wanpeng Zhang, Sipeng Zheng, Ziheng Xi, Chaoyi Xu, Haiweng Xu, Haoqi Yuan, Chi Zhang, Yiqing Wang, et al. Being-h0. 5: Scaling human-centric robot learning for cross-embodiment generalization.arXiv preprint arXiv:2601.12993, 2026

    2026

  16. [24]

    Green-vla: Staged vision-language-action model for generalist robots

    I Apanasevich, M Artemyev, R Babakyan, P Fedotova, D Grankin, E Kupryashin, A Misailidi, D Nerus, A Nutalapati, G Sidorov, et al. Green-vla: Staged vision-language-action model for generalist robots. arXiv preprint arXiv:2602.00919, 2026

    2026

  17. [25]

    Hex: Humanoid-aligned experts for cross-embodiment whole-body manipulation.arXiv preprint arXiv:2604.07993, 2026

    Shuanghao Bai, Meng Li, Xinyuan Lv, Jiawei Wang, Xinhua Wang, Fei Liao, Chengkai Hou, Langzhe Gu, Wanqi Zhou, Kun Wu, et al. Hex: Humanoid-aligned experts for cross-embodiment whole-body manipulation.arXiv preprint arXiv:2604.07993, 2026

    2026 arXiv

  18. [26]

    Hamster: Hierarchicalactionmodelsforopen-worldrobotmanipulation

    Yi Li, Yuquan Deng, Jesse Zhang, Joel Jang, Marius Memmel, Caelan Garrett, Fabio Ramos, Dieter Fox, AnqiLi,AbhishekGupta,etal. Hamster: Hierarchicalactionmodelsforopen-worldrobotmanipulation. InInternational Conference on Learning Representations, volume 2025, pages 24040–24068, 2025

    2025

  19. [27]

    Robotic control via embodied chain-of-thought reasoning.arXiv preprint arXiv:2407.08693, 2024

    Michał Zawalski, William Chen, Karl Pertsch, Oier Mees, Chelsea Finn, and Sergey Levine. Robotic control via embodied chain-of-thought reasoning.arXiv preprint arXiv:2407.08693, 2024

    2024 arXiv

  20. [28]

    Emma-x: An embodied multimodal action model with grounded chain of thought and look-ahead spatial reasoning

    Qi Sun, Pengfei Hong, Tej Deep Pala, Vernon Toh, U-Xuan Tan, Deepanway Ghosal, and Soujanya Poria. Emma-x: An embodied multimodal action model with grounded chain of thought and look-ahead spatial reasoning. InProceedings of the 63rd Annual Meeting of the Association for Compu...

    2025

  21. [29]

    Tracevla: Visual trace prompting enhances spatial-temporal awareness for generalist robotic policies

    Ruijie Zheng, Yongyuan Liang, Shuaiyi Huang, Jianfeng Gao, Hal Daumé III, Andrey Kolobov, Furong Huang, and Jianwei Yang. Tracevla: Visual trace prompting enhances spatial-temporal awareness for generalist robotic policies. InInternational Conference on Learning Representations...

    2025

  22. [30]

    Qwen Team. Qwen3. 5: Towards native multimodal agents, february 2026.URL https://qwen. ai/blog, 2026

    2026

  23. [31]

    Minivla: A better vla with a smaller footprint, 2024

    Suneel Belkhale and Dorsa Sadigh. Minivla: A better vla with a smaller footprint, 2024. URL https://github.com/Stanford-ILIAD/openvla-mini

    2024

  24. [32]

    Actioncodec: What makes for good action tokenizers.arXiv preprint arXiv:2602.15397, 2026

    Zibin Dong, Yicheng Liu, Shiduo Zhang, Baijun Ye, Yifu Yuan, Fei Ni, Jingjing Gong, Xipeng Qiu, Hang Zhao, Yinchuan Li, and Jianye Hao. Actioncodec: What makes for good action tokenizers.arXiv preprint arXiv:2602.15397, 2026

    2026

  25. [33]

    FASTer: Toward powerful and efficient autoregressive vision–language–action models with learnableactiontokenizerandblock-wisedecoding

    Yicheng Liu, Shiduo Zhang, Zibin Dong, Baijun Ye, Tianyuan Yuan, Xiaopeng Yu, Linqi Yin, Chenhao Lu, Junhao Shi, Luca Jiang-Tao Yu, Liangtao Zheng, Jingjing Gong, Tao Jiang, Xipeng Qiu, and Hang Zhao. FASTer: Toward powerful and efficient autoregressive vision–language–action m...

    2026

  26. [34]

    Physical Intelligence, Bo Ai, Ali Amin, Raichelle Aniceto, Ashwin Balakrishna, Greg Balke, Kevin Black, George Bokinsky, Shihao Cao, Thomas Charbonnier, et al.𝜋0.7: a steerable generalist robotic foundation model with emergent capabilities.arXiv preprint arXiv:2604.15483, 2026

    2026 arXiv

  27. [35]

    Gemini 3 pro model card, 2026

    Google DeepMind. Gemini 3 pro model card, 2026. URL https://storage.googleapis.com/ deepmind-media/Model-Cards/Gemini-3-Pro-Model-Card.pdf . Model card, last updated May 2026

    2026

  28. [36]

    Seed 2.0 official launch, 2026

    ByteDance Seed Team. Seed 2.0 official launch, 2026. URLhttps://research.doubao.com/en/blog/ seed-2-0-official-launch

    2026

  29. [37]

    Sam 3: Segment anything with concepts, 2025

    Nicolas Carion, Laura Gustafson, Yuan-Ting Hu, Shoubhik Debnath, Ronghang Hu, Didac Suris, Chaitanya Ryali, Kalyan Vasudev Alwala, Haitham Khedr, Andrew Huang, Jie Lei, Tengyu Ma, Baishan Guo, Arpit Kalla, Markus Marks, Joseph Greer, Meng Wang, Peize Sun, Roman Radle, Triantaf...

    2025 arXiv

  30. [38]

    Llava-next: Improved reasoning, ocr, and world knowledge, January 2024

    HaotianLiu,ChunyuanLi,YuhengLi,BoLi,YuanhanZhang,ShengShen,andYongJaeLee. Llava-next: Improved reasoning, ocr, and world knowledge, January 2024. URLhttps://llava-vl.github.io/ blog/2024-01-30-llava-next/

    2024

  31. [39]

    Llava-onevision: Easy visual task transfer, 2024

    Bo Li, Yuanhan Zhang, Dong Guo, Renrui Zhang, Feng Li, Hao Zhang, Kaichen Zhang, Peiyuan Zhang, Yanwei Li, Ziwei Liu, and Chunyuan Li. Llava-onevision: Easy visual task transfer, 2024. URL https://arxiv.org/abs/2408.03326

    2024 arXiv

  32. [40]

    Robopoint: A vision-language model for spatial affordance prediction for robotics, 2024

    Wentao Yuan, Jiafei Duan, Valts Blukis, Wilbert Pumacay, Ranjay Krishna, Adithyavairavan Murali, Arsalan Mousavian, and Dieter Fox. Robopoint: A vision-language model for spatial affordance prediction for robotics, 2024. URLhttps://arxiv.org/abs/2406.10721

    2024 arXiv

  33. [41]

    Molmoact: Action reasoning models that can reason in space, 2025

    Jason Lee, Jiafei Duan, Haoquan Fang, Yuquan Deng, Shuo Liu, Boyang Li, Bohan Fang, Jieyu Zhang, Yi Ru Wang, Sangho Lee, Winson Han, Wilbert Pumacay, Angelica Wu, Rose Hendrix, Karen Farley, Eli VanderBilt, Ali Farhadi, Dieter Fox, and Ranjay Krishna. Molmoact: Action reasonin...

    2025 arXiv

  34. [42]

    Robobrain: A unified brain model for robotic manipulation from abstract to concrete, 2025

    Yuheng Ji, Huajie Tan, Jiayu Shi, Xiaoshuai Hao, Yuan Zhang, Hengyuan Zhang, Pengwei Wang, MengdiZhao, YaoMu, PengjuAn, XindaXue, QinghangSu, HuaihaiLyu, XiaolongZheng, JiamingLiu, Zhongyuan Wang, and Shanghang Zhang. Robobrain: A unified brain model for robotic manipulation f...

    2025 arXiv

  35. [43]

    DROID: A large-scale in-the-wild robot manipulation dataset.arXiv preprint arXiv:2403.12945, 2024

    Alexander Khazatsky, Karl Pertsch, Suraj Nair, Ashwin 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....

    2024 arXiv

  36. [44]

    Molmoact2: Action reasoning models for real-world deployment, 2026

    Haoquan Fang, Jiafei Duan, Donovan Clay, Sam Wang, Shuo Liu, Weikai Huang, Xiang Fan, Wei- Chuan Tsai, Shirui Chen, Yi Ru Wang, Shanli Xing, Jaemin Cho, Jae Sung Park, Ainaz Eftekhar, Peter Sushko, Karen Farley, Angad Wadhwa, Cole Harrison, Winson Han, Ying-Chun Lee, Eli Vande...

    2026 arXiv

  37. [45]

    Bridgedata v2: A dataset for robot learning at scale

    Homer Rich Walke, Kevin Black, Tony Z Zhao, Quan Vuong, Chongyi Zheng, Philippe Hansen-Estruch, Andre 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

    2023

  38. [46]

    Starvla: A lego-like codebase for vision- language-action model developing, 2026

    StarVLA Community and Von Neumann Institute, HKUST. Starvla: A lego-like codebase for vision- language-action model developing, 2026

    2026

  39. [47]

    Memoryvla: Perceptual-cognitive memory in vision-language-action models for robotic manipulation, 2025

    Hao Shi, Bin Xie, Yingfei Liu, Lin Sun, Fengrong Liu, Tiancai Wang, Erjin Zhou, Haoqiang Fan, Xiangyu Zhang, and Gao Huang. Memoryvla: Perceptual-cognitive memory in vision-language-action models for robotic manipulation, 2025

    2025

  40. [48]

    Eo-1: An open unified embodied foundation model for general robot control.arXiv preprint arXiv:2508.21112, 2026

    Delin Qu, Haoming Song, Qizhi Chen, Zhaoqing Chen, Xianqiang Gao, Dong Wang, Xinyi Ye, Qi Lv, Modi Shi, Guanghui Ren, Cheng Ruan, Maoqing Yao, Haoran Yang, Jiacheng Bao, Bin Zhao, and Xuelong Li. Eo-1: An open unified embodied foundation model for general robot control.arXiv p...

    2026

  41. [49]

    Xiaomi-robotics-0: An open-sourced vision-language-action model with real-time execution.arXiv preprint arXiv:2602.12684, 2026

    Rui Cai, Jun Guo, Xinze He, Piaopiao Jin, Jie Li, Bingxuan Lin, Futeng Liu, Wei Liu, Fei Ma, Kun Ma, Feng Qiu, Heng Qu, Yifei Su, Qiao Sun, Dong Wang, Donghao Wang, Yunhong Wang, Rujie Wu, Diyun Xiang, Yu Yang, Hangjun Ye, Yuan Zhang, and Quanyun Zhou. Xiaomi-robotics-0: An op...

    2026

  42. [50]

    Motus: A unified latent action world model.arXiv preprint arXiv:2512.13030, 2025

    Hongzhe Bi, Hengkai Tan, Shenghao Xie, Zeyuan Wang, Shuhe Huang, Haitian Liu, Ruowen Zhao, Yao Feng, Chendong Xiang, Yinze Rong, Hongyan Zhao, Hanyu Liu, Zhizhong Su, Lei Ma, Hang Su, and Jun Zhu. Motus: A unified latent action world model.arXiv preprint arXiv:2512.13030, 2025

    2025 arXiv

  43. [51]

    Causal world modeling for robot control.arXiv preprint arXiv:2601.21998, 2026

    Lin Li, Qihang Zhang, Yiming Luo, Shuai Yang, Ruilin Wang, Fei Han, Mingrui Yu, Zelin Gao, Nan Xue, Xing Zhu, Yujun Shen, and Yinghao Xu. Causal world modeling for robot control.arXiv preprint arXiv:2601.21998, 2026

    2026 arXiv

  44. [52]

    A pragmatic vla foundation model.arXiv preprint arXiv:2601.18692, 2026

    Wei Wu, Fan Lu, Yunnan Wang, Shuai Yang, Shi Liu, Fangjing Wang, Qian Zhu, He Sun, Yong Wang, Shuailei Ma, Yiyu Ren, Kejia Zhang, Hui Yu, Jingmei Zhao, Shuai Zhou, Zhenqi Qiu, Houlong Xiong, ZiyuWang,ZechenWang,RanCheng,Yong-LuLi,YongtaoHuang,XingZhu,YujunShen,andKecheng Zheng...

    2026 arXiv

  45. [53]

    Qwen-vla: Unifying vision-language-action modeling across tasks, environments, and robot embodiments.arXiv preprint arXiv:2605.30280, 2026

    Qwen Team. Qwen-vla: Unifying vision-language-action modeling across tasks, environments, and robot embodiments.arXiv preprint arXiv:2605.30280, 2026

    2026 arXiv

  46. [54]

    Internvla-m1: A spatially guided vision-language-action framework for generalist robot policy.arXiv preprint arXiv:2510.13778, 2025

    Xinyi Chen, Yilun Chen, Yanwei Fu, Ning Gao, Jiaya Jia, Weiyang Jin, Hao Li, Yao Mu, Jiangmiao Pang, Yu Qiao, Yang Tian, Bin Wang, Bolun Wang, Fangjing Wang, Hanqing Wang, Tai Wang, Ziqin Wang, Xueyuan Wei, Chao Wu, Shuai Yang, Jinhui Ye, Junqiu Yu, Jia Zeng, Jingjing Zhang, J...

    arXiv

  47. [55]

    Igniting vlms toward the embodied space.arXiv preprint arXiv:2509.11766, 2025

    Andy Zhai, Brae Liu, Bruno Fang, Chalse Cai, Ellie Ma, Ethan Yin, Hao Wang, Hugo Zhou, James Wang, Lights Shi, Lucy Liang, Make Wang, Qian Wang, Roy Gan, Ryan Yu, Shalfun Li, Starrick Liu, Sylas Chen, Vincent Chen, and Zach Xu. Igniting vlms toward the embodied space.arXiv pre...

    2025

  48. [56]

    Fine-tuning vision-language-action models: Optimizing speed and success.arXiv preprint arXiv:2502.19645, 2025

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

    2025 arXiv

  49. [57]

    Cosmos policy: Fine-tuning video models for visuomotor control and planning.arXiv preprint arXiv:2601.16163, 2026

    MooJinKim,YihuaiGao,Tsung-YiLin,Yen-ChenLin,YunhaoGe,GraceLam,PercyLiang,ShuranSong, Ming-Yu Liu, Chelsea Finn, and Jinwei Gu. Cosmos policy: Fine-tuning video models for visuomotor control and planning.arXiv preprint arXiv:2601.16163, 2026

    2026 arXiv

  50. [58]

    Task adaptation of vision-language-action model: 1st place solution for the 2025 behavior challenge.arXiv preprint arXiv:2512.06951, 2025

    Ilia Larchenko, Gleb Zarin, and Akash Karnatak. Task adaptation of vision-language-action model: 1st place solution for the 2025 behavior challenge.arXiv preprint arXiv:2512.06951, 2025

    2025

  51. [59]

    Openpi comet: Competition solution for 2025 behavior challenge.arXiv preprint arXiv:2512.10071, 2025

    Junjie Bai, Yu-Wei Chao, Qizhi Chen, Jinwei Gu, Moo Jin Kim, Zhaoshuo Li, Xuan Li, Tsung-Yi Lin, Ming-Yu Liu, Nic Ma, et al. Openpi comet: Competition solution for 2025 behavior challenge.arXiv preprint arXiv:2512.10071, 2025

    2025

  52. [60]

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300, 2024

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, et al. Deepseekmath: Pushing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300, 2024

    2024 arXiv

  53. [61]

    RLinf: Flexible and efficient large-scale reinforcement learning via macro-to-micro flow transformation.arXiv preprint arXiv:2509.15965, 2025

    Chao Yu, Yuanqing Wang, Zhen Guo, Hao Lin, Si Xu, Hongzhi Zang, Quanlu Zhang, Yongji Wu, Chunyang Zhu, Junhao Hu, et al. RLinf: Flexible and efficient large-scale reinforcement learning via macro-to-micro flow transformation.arXiv preprint arXiv:2509.15965, 2025. 30 A Appendix...

    2025

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.