Pith.

Robotics · arXiv:2610.18774 v1 · submitted 13 Oct 2026 · 107 references

GreenWAM: A World Foundation Model for Robotics

Gennady Sidorov, Valerii Kornilov, Denis Grankin, Polina Fedotova, Aleksey Postnikov

The paper says

“A robot needs a world model that predicts how its actions will change the scene and a policy that chooses those actions.”

more of the abstract

Models are usually built and evaluated for only one of these roles, and every robot records its actions in its own format. We ask whether one robot world model can serve both roles across different bodies, and what makes this work. We present GreenWAM, a world foundation model for robotics, together with the recipe that produces it: (1) a shared anatomical action space that describes 20 robot families, from bimanual arms to dexterous humanoids, in common joint coordinates; (2) joint training on these families in visual, action-conditioned and world-action modes; and (3) policies that split an action expert off the generator. As a world model, GreenWAM attains top-tier trajectory, depth and 3D accuracy on WorldArena 1.0, ahead of every world-action model and video generator in our comparison. As a policy, it scores among the top entries of the RoboTwin 2.0 leaderboard, reaches the highest zero-shot LIBERO-plus success among the compared methods and is competitive with dedicated vision-language-action and world-action policies on CALVIN and LIBERO. We further adapt GreenWAM to a real humanoid, where it succeeds more often than policies adapted from π0.5 and from Cosmos 3 Edge, and distill it into a single-step policy for real-time control. Controlled experiments yield three findings: robot specialization improves world prediction and control together; an action expert helps when split off at the start of adaptation; and keeping world-model modes during policy adaptation preserves robustness to scene changes.

from the abstract

Pith says

Machine review · deepseek-v4-flash · 14 Oct 2026 · what was read was not recorded · not checked by a person

It can be wrong. Challenge this review or reply to it.

Holds up, with conditions. A strong, unusually honest systems paper on a shared world-model/policy foundation, with the main open risk being the breadth of anatomical-codec validation and single-run benchmark numbers. This line is wrong

2 major objections · 5 minor

How often objections from reviews like this one are right: not yet measured

Are you an author? What you can do.

Not checked: figures, code, what the 107 references say.

The paper says, at A.3; Eq. (1); B.5

The review did not quote this passage.

Pith says: Objection 1 of 2 · major

The shared anatomical action space is the load-bearing premise for every cross-embodiment claim, yet Section A.3 states that forward-kinematics validation of the codecs against measured end-effector poses is performed only 'for the main families', without defining which families these are, while 43 source layouts are registered and Figure 27 shows filled slots for all 20 families.

more of objection 1

A wrong joint index, sign, zero, or per-layout normalization statistic in an unvalidated codec would send corrupted targets through the single shared projector for that body, so the held-out six-family evaluations of Section 5.1 and the breadth of the '20 families' claim would not follow from the stated mechanism. I acknowledge that the paper explicitly discloses the partial validation, and that the headline WorldArena, RoboTwin and RoboDojo results rest on the Worker/ARX5 codec, whose reported forward-kinematics reproduction error (5.2 and 7.1 mm) is reassuring; the gap is therefore local rather than fatal. Please (i) enumerate the FK-validated families and report their end-effector reproduction errors, and (ii) for the CALVIN/LIBERO single-arm Franka results, which enter through the Cartesian end-effector slots with an unchecked SE(3) frame convention (Section B.5), provide a validation of that convention against ground-truth poses in the LIBERO simulator.

The paper says, at 6.5; Fig. 17; C.2; Table 14

The review did not quote this passage.

Pith says: Objection 2 of 2 · major

The abstract claims GreenWAM 'succeeds more often' than the Cosmos 3 Edge policy and pi0.5 on the real humanoid, but Section C.2 shows this rests on 43/45 versus 40/45 versus 38/45 in-distribution instructions and 21/25 versus 20/25 versus 19/25 out-of-distribution instructions; those differences are within the sampling uncertainty of a matched binary design of this size.

more of objection 2

The protocol is transparent and the sample is typical for real-robot work, but unlike the RoboTwin comparisons of Section C.4 no interval is reported for the humanoid result; because the design is matched by scene configuration, a paired bootstrap over configurations is feasible. The same single-run caveat applies to the LIBERO-plus headline, where 86.8% versus 86.1% for CKT-WAM and 85.7% for pi0.5 (Table 14) is a 0.7-point margin. Please report intervals for these small-margin claims or temper the wording in the abstract.

5 minor objections

The paper says, at Section 5.1

Pith says: minor objection 1 of 5

The action-following evaluation uses one held-out recording per family and reports no per-family uncertainty; the summary statistics ('less than a fifth' of the motion under zero command, 'about a quarter more' at 1.5x) would be easier to assess with per-family values and a variance estimate.

The paper says, at Section 5.2

Pith says: minor objection 2 of 5

The margin over the compared world-action models under the motion-capped EWMScore is small (63.31 versus 63.16 and 62.99, both upper bounds for competitors), so a sensitivity check in which the motion cap is applied to GreenWAM at the leaderboard-mean level as well would clarify how much of the ordering depends on the cap convention.

show 3 more

The paper says, at Section 3.2

Pith says: minor objection 3 of 5

The statement that 'no robot has action parameters of its own' should be reconciled with the per-layout normalization statistics mu and s in Equation (1), which are layout-specific inputs to the codec; phrasing such as 'no trainable action parameters in the model' would be more precise.

The paper says, at Section 4.3

Pith says: minor objection 4 of 5

The assertion that video and state 'carry equal weight' because both terms are scaled by lambda = 10 is nominal, since the two terms have different scales; a sentence on how the balance was chosen would aid reproducibility.

The paper says, at Throughout

Pith says: minor objection 5 of 5

Rendering artifacts such as 'official' written with a ligature-ff and the spaced 'GreenW AM' recur throughout the manuscript; they do not affect the technical content but should be fixed in the camera-ready version.

What would settle it

Run the Section A.3 forward-kinematics check on all 43 registered layouts, not only the main families: for held-out episodes with both native joint states and measured end-effector poses, apply each layout's codec (Equation 1) to obtain slot values, compute the predicted end-effector pose from the layout URDF, and compare with the measured pose at each time step. Report per-layout median and maximum translation and orientation error, using the Worker/ARX5 result (5.2-7.1 mm) as the tolerance anchor. If every layout passes at comparable tolerance, the shared-space premise is supported; if any layout fails, the downstream WorldArena, RoboTwin, CALVIN or LIBERO conclusions that rely on that codec should be rerun with the codec corrected, or the affected family's results should be qualified.

from the skeptic's reading

How the review went

4 readings by deepseek-v4-flash · 14 Oct 2026

  1. Reader holds up, with conditions confidence moderate
  2. Skeptic agrees 1 min 39 s · 52,069 tokens read, 13,698 written
  3. Editor wrote the verdict above 22 s · 52,983 tokens read, 2,557 written
  4. Referee major revision: 2 major objections, 5 minor 2 min 18 s · 52,453 tokens read, 22,381 written

Not run on this paper: the simulated authors' defence.

Pith did not check this review's objections against the paper.

The review in full

The review in brief, machine-written

Short version: this is one of the more careful world-action model papers I have read from an industrial group. The 79-slot anatomical action space, the seven-mode joint training, and the action-expert split are a real integration, not a repackaging. The authors do the right work around it: whole-subtask atoms, a disclosed data mixture, held-out prediction, paired controlled comparisons with hierarchical bootstrap intervals, external leaderboards, a real humanoid evaluation, and a distilled policy. The three controlled findings — specialization helps both roles, the expert split helps, and retaining world-model modes helps randomized scenes — are empirical comparisons rather than fits, and they hold up on reading.

The main soft spot is the one the stress-test flags, in weaker form. Section A.3 validates forward kinematics against measured end-effector poses only for the 'main families' and does not define that set, while 43 codecs are registered. A wrong sign, zero, slot, or normalization statistic in an unvalidated codec would corrupt that family's targets as they pass through the shared projector. The paper discloses the partial validation and keeps unresolved channels out of the shared slots, and the body behind the headline benchmark, Worker/ARX5, is validated, so the central results do not collapse. But the '20 families' claim is broader than the verification, and the CALVIN/LIBERO Franka result depends on the Cartesian end-effector slot convention without an independently checked SE(3) frame. I would call that a moderate reproducibility gap, not a fatal flaw.

The other soft spots are smaller. The WorldArena comparison uses a disclosed but non-official motion-capped aggregate; competitors receive an upper bound, which is fair but makes the ranking approximate. The RoboTwin leaderboard numbers are single runs without uncertainty, and the humanoid comparison covers 70 instructions per policy, so a few trials separate the methods. RoboTwin also shares the ARX5 body with the pretraining Worker/ARX5 family; this is disclosed and not circular, but it means RoboTwin does not demonstrate cross-embodiment transfer the way the Franka and humanoid results do. The full training corpus and base checkpoint are not released; policy and world-model weights plus evaluation code are, which is a reasonable middle ground.

This paper is for anyone building or benchmarking world-action models. It deserves a serious referee; the requested fixes are bounded: define the validated codec set, report uncertainty on headline numbers, and clarify the WorldArena aggregate. I would engage with it.

The strongest objection

The central claim requires Equation (1)'s premise: one slot in the 79-slot anatomical space "means the same joint motion on every robot" (Section 3.2). This is what lets one shared projector serve all bodies, and both the world-model and policy results inherit it. Section A.3 states that correspondences are established from the URDF at zero pose, manufacturer documentation or named channels, and that "for the main families the joint assignment is validated by forward kinematics against measured end-effector poses." The set of "main families" is not defined, and 43 source layouts are registered; Figures 4 and 27 show many families whose arm, hand and body slots are filled without a reported FK check. A wrong joint index, sign, zero, slot, or per-layout normalization statistic in any unvalidated codec would put corrupted targets through the single shared projector for that body, so the cross-embodiment transfer results for that family would no longer follow from the stated mechanism. The CALVIN/LIBERO Franka result depends on the same interface through Cartesian end-effector slots (Section B.5), where the SE(3) frame convention is not independently checked. The paper discloses the partial nature of the validation, and the Worker/ARX5 FK check in Section A.3 is reassuring for the main benchmark, but the breadth of the "20 families" claim rests on more codecs than are validated.

The reader's CONDITIONAL verdict already identifies the anatomical action space as the weakest assumption, and my read agrees. I considered the WorldArena motion-capped aggregate, the RoboTwin Worker/ARX5 codec reuse, and the small real-robot trial counts; all are disclosed and weigh on confidence, but none is as load-bearing as the codec validation. If a codec silently swaps, flips, or re-zeroes a joint before it enters the single shared projector, every mode in Equation (3) is trained on wrong targets for that layout, and the central "one slot means one joint motion" premise fails. That premise is the mechanism that distinguishes GreenWAM from per-embodiment projections such as Cosmos 3 Edge and GE-Sim 2.0, and it underwrites the 20-family breadth of the claim. The paper gives partial support: the Worker/ARX5 FK check reproduces RoboTwin gripper paths to 5.2-7.1 mm, and the main families are validated. But "main families" is unspecified, and Section A.3 explicitly leaves the remaining layouts to URDF, documentation and named-channel correspondence. The proposed check is feasible with the released code and would settle whether the concern is real. Because this is a conditional risk that the paper has disclosed, the reader's CONDITIONAL verdict stands; I would not change it.

The referee's report

The contribution is potentially significant for robotics foundation models: one system demonstrates that a video world foundation can be adapted to many embodiments through a single shared action interface, that the same checkpoint serves both as an action-conditioned world model and as a policy, and that three concrete design choices - robot-specialized initialization, an early action-expert split, and retained world-model modes - each improve downstream results at a fixed budget. The evidence base is above the field's norm: external leaderboards with published protocols (WorldArena 1.0, RoboTwin 2.0, RoboDojo), held-out and real-robot evaluations, released weights and evaluation code, a detailed training and sampling recipe, and paired comparisons with hierarchical bootstrap intervals. The WorldArena comparison is conservative toward competitors, whose scores are upper bounds under the motion cap, and the paper candidly states its limitations, including sensitivity to task geometry, autoregressive error accumulation, per-benchmark post-training, and the partial forward-kinematics validation of the codecs.

Pith's reading of the paper (a machine's summary, not the authors')

Robots today are split into two kinds of models: ones that predict how a scene will change and ones that choose actions, and each robot records its actions in its own format. GreenWAM asks whether a single world foundation model can do both jobs across different bodies, and what makes that possible. The paper's answer is yes, provided the robot's body is described in a shared anatomical coordinate system: 79 slots covering legs, torso, arms, and hands, in which one slot means the same joint motion on every robot. Retrained from Cosmos 3 Edge on 20 robot families and trained jointly in seven video-prediction, action-conditioned, and world-action modes, the same checkpoint predicts future video and geometry accurately on WorldArena 1.0, scores among the top RoboTwin 2.0 policies, and beats policies adapted from π0.5 and from its own parent on a real humanoid. If this holds, one foundation can replace the usual one-model-per-robot, one-role-per-model arrangement.

What carries the argument. The load-bearing object is the shared anatomical action space: 79 slots in four blocks (12 legs, 23 body, 18 arms, 26 hands) into which layout-specific codecs place each robot's native state channels, fixing sign and zero per slot and masking absent coordinates. The mapping is defined by $q_j = \rho_{\ell j}(x_{\pi_\ell(j)} - o_{\ell j})$ with a validity mask $m_{\ell j}$, and relative targets are decoded as offsets from the block-boundary state $q_0$ by $\hat{q}_k = \hat{\delta}_k + d_\ell \odot q_0$. This one mapping is what lets a single trained input/output projector serve all 20 families, and it is why a new robot needs only a new codec rather than new action parameters. Around that interface, the robotics stage trains seven modes with one rectified-flow objective and a shared noise level, so video and joint-state prediction are learned by the same weights. Policy adaptation then copies the generator tower into an action expert whose feed-forward blocks diverge while attention spans both streams, and a consistency-distillation recipe collapses the multi-step sampler into a one-step student that answers in 35.4 ms.

Core claim. The central claim is that one checkpoint serves both roles—world model and policy—across 20 robot families, from bimanual arms to humanoids with dexterous hands, through a single shared action interface. The paper argues that this works because the interface is anatomical rather than robot-specific: layout codecs map each robot's native joints into 79 shared slots with fixed sign and zero conventions, so one slot means the same joint motion on every body, and a single shared projector handles all of them. Jointly training on seven modes—text-to-video, image-to-video with and without history, forward dynamics with and without history, inverse dynamics, and world-action prediction—gives one set of weights the ability to generate video, follow a commanded trajectory, and emit state trajectories from video. As a policy, the foundation splits a copy of its generator tower into a dedicated action expert whose parameters diverge while joint attention keeps it coupled to the video stream, preserving the initial function while giving actions their own capacity. The reported outcomes are top trajectory, depth and 3D accuracy on WorldArena 1.0, a 67.5% average on the RoboTwin 2.0 leaderboard, 4.64 completed instructions on CALVIN, 97.4% on LIBERO, 86.8% zero-shot on LIBERO-plus, and 95.6% in-distribution success on a real humanoid, ahead of policies adapted from π0.5 and from Cosmos 3 Edge. Controlled comparisons attribute the gains to three choices: robot-specialized initialization, an action expert split off at the start of adaptation, and keeping small shares of world-model modes during policy adaptation.

The premise it rests on. The load-bearing premise is that the 79-slot anatomical action space really means the same joint motion on every robot: if any layout's codec assigns the wrong joint, sign, zero, or slot, the shared projector is trained on corrupted targets for that body and that family's results no longer follow, while forward-kinematics validation against measured poses is described only for the main families.

If the paper is right (Pith's inferences, not the authors')

A single checkpoint, not a zoo of per-robot models, can supply both forward prediction and control; the paper demonstrates this on six held-out families for video prediction and on four simulation benchmarks plus a real humanoid for control.

New robot bodies plug into the foundation through a layout codec that maps their joints onto the existing 79 slots, without training new action projections; the paper shows a Franka arm, absent from its corpus, entering through the Cartesian end-effector slots.

Robot specialization is not just a policy trick: starting from the robot-specialized checkpoint improved trajectory accuracy on WorldArena by 3.0 points and closed-loop success on RoboTwin by 5.4 to 8.4 points at a fixed adaptation budget.

Keeping video-prediction modes during policy adaptation raised randomized-scene success on RoboTwin by 8.0%, and decoding the policy's own predictions shows that out-of-distribution failures track the accuracy of the imagined scene, while in-distribution failures are action errors despite correct imagination.

The distilled one-step student keeps most of the teacher's success on the real humanoid (91.1% versus 95.6% in distribution) while answering 12.3 times faster, at 28.3 Hz.

If the anatomical slot correspondence transfers as cleanly as claimed, the same interface should support few-shot control transfer across bodies that share arm kinematics without per-pair alignment; the paper does not test this directly, but its codec-only adaptation story predicts it.

The WorldArena motion-cap analysis suggests a general evaluation hazard: many video generators are rewarded for moving more than the recorded robot, so world-model rankings can invert once predictions must match commanded actions; benchmark designs that enforce physical-action fidelity are likely to propagate.

A testable extension is motion-level data augmentation: the paper's appearance augmentations add no new trajectories, and a 3 cm table-height change still costs 8% success, so synthesizing height and geometry variations in the recorded states is a direct next experiment suggested by the paper's own limitation analysis.

The correlation between imagination accuracy and success on randomized RoboTwin scenes implies that world-model prediction error could serve as a policy selection or early-stopping signal in simulation, which the paper lists as an outlook for its world-model role.

Check this review

Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.

How often reviews like this one are right

Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.

How often objections from reviews like this one are right: not yet measured

Every value above, in the review's public record (JSON)

Reviews by people

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review

How Pith checks papers · The full audit of this paper

The paper

Abstract

A robot needs a world model that predicts how its actions will change the scene and a policy that chooses those actions. Models are usually built and evaluated for only one of these roles, and every robot records its actions in its own format. We ask whether one robot world model can serve both roles across different bodies, and what makes this work. We present GreenWAM, a world foundation model for robotics, together with the recipe that produces it: (1) a shared anatomical action space that describes 20 robot families, from bimanual arms to dexterous humanoids, in common joint coordinates; (2) joint training on these families in visual, action-conditioned and world-action modes; and (3) policies that split an action expert off the generator. As a world model, GreenWAM attains top-tier trajectory, depth and 3D accuracy on WorldArena 1.0, ahead of every world-action model and video generator in our comparison. As a policy, it scores among the top entries of the RoboTwin 2.0 leaderboard, reaches the highest zero-shot LIBERO-plus success among the compared methods and is competitive with dedicated vision-language-action and world-action policies on CALVIN and LIBERO. We further adapt GreenWAM to a real humanoid, where it succeeds more often than policies adapted from π0.5 and from Cosmos 3 Edge, and distill it into a single-step policy for real-time control. Controlled experiments yield three findings: robot specialization improves world prediction and control together; an action expert helps when split off at the start of adaptation; and keeping world-model modes during policy adaptation preserves robustness to scene changes.

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.

abstract compiled · 2026-10-14

The authors' main claim, as the classifier picked it out 0.87

“Controlled experiments yield three findings: robot specialization improves world prediction and control together; an action expert helps when split off at the start of adaptation; and keeping world-model modes during policy adaptation preserves robustness to scene changes.”

How the main claim is stated: asserted 1.00

What the abstract itself states, as probabilities: claims first 0.12 · states limitation 0.19 · human participants 0.04 · controlled comparison 0.84 · randomized assignment 0.02 · states quantitative result 0.50 · compares with prior results 0.98 · mentions code or data availability 0.05

The abstract, sentence by sentence

This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.

SentenceRoleThe authors' own workStates the main conclusionStrength
A robot needs a world model that predicts how its actions will change the scene and a policy that chooses those actions. background 1.00 0.05 0.02 asserted 1.00
Models are usually built and evaluated for only one of these roles, and every robot records its actions in its own format. gap 0.81 0.07 0.02 asserted 0.99
We ask whether one robot world model can serve both roles across different bodies, and what makes this work. gap 0.85 0.12 0.03 unclear 0.88
We present GreenWAM, a world foundation model for robotics, together with the recipe that produces it: (1) a shared anatomical action space that describes 20 robot families, from bimanual arms to dexterous humanoids, in common joint coordinates; (2) joint training on these families in visual, action-conditioned and world-action modes; and (3) policies that split an action expert off the generator. contribution 0.59 0.95 0.12 asserted 1.00
As a world model, GreenWAM attains top-tier trajectory, depth and 3D accuracy on WorldArena 1.0, ahead of every world-action model and video generator in our comparison. result 1.00 0.95 0.17 asserted 1.00
As a policy, it scores among the top entries of the RoboTwin 2.0 leaderboard, reaches the highest zero-shot LIBERO-plus success among the compared methods and is competitive with dedicated vision-language-action and world-action policies on CALVIN and LIBERO. result 1.00 0.95 0.17 asserted 1.00
We further adapt GreenWAM to a real humanoid, where it succeeds more often than policies adapted from $\pi_{0.5}$ and from Cosmos 3 Edge, and distill it into a single-step policy for real-time control. result 0.93 0.96 0.10 asserted 1.00
Controlled experiments yield three findings: robot specialization improves world prediction and control together; an action expert helps when split off at the start of adaptation; and keeping world-model modes during policy adaptation preserves robustness to scene changes. result 0.90 0.97 0.47 asserted 1.00

What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 0.90

What was examined, and what was not

Sources read

  • arXiv listing abstract v1

Not examined

  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers

Levels run

  • P abstract

Not run

  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

How the labels were made

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-10-14: 1 request, 6,534 input tokens.

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON

Figures (16)
Figure 3 of the paper
Figure 3. GreenWAM architecture, shown for the world–action (WAM) mode on one held-out Green humanoid episode. The task instruction, the multi-camera canvas and the robot’s native joint state at 𝑡 = 0 enter a frozen reasoner, a trained generator for the video tokens and, in the adapted policy, an action expert for the state and action tokens, all sharing multimodal attention (the Cosmos 3 Edge mixture of transformers); the layout’s codec maps the native joints to the 79 anatomical slots before the shared input projection (
Figure 16 of the paper
Figure 16. GreenWAM policy inference on the Green humanoid. Eight frames of one continuous episode, cropped from a side camera: the robot puts a cap and a bottle into a bag, lifts the bag by its handles and hands it to a person. Colors mark the three subtasks. We evaluate the policies on a packing task in which the Green humanoid places a bottle and a cap into a bag, then lifts the loaded bag with both hands. For evaluation the task is given as five consecutive instructions, grasping and placing each object and then lifting the bag; each names the object and, for the single-arm actions, the…

Showing 2 of the paper's 16 figures. Figures are the authors', reproduced from the paper.

References (107)
  1. [1] 1X Technologies. 1X World Model: Evaluating bits, not atoms. Technical progress report, 2025. URL https://www.1x.tech/ 1x-world-model.pdf on Pith
  2. [2] AgiBot Research Team, R. Liu, W. Zhao, Z. Yang, L. Chen, P. Zhou, S. Chen, G. Ren, Y. Peng, et al. GE-Act 2.0: Pretraining and scaling a world-action model for robotic manipulation, 2026. URL https://arxiv.org/abs/2609.05588 on PithPith's reviewarXiv
  3. [3] AgiBot World Team. AgiBot World 2026. Hugging Face dataset, 2026. URL https://huggingface.co/datasets/agibot-world/ AgiBotWorld2026 on Pith
  4. [4] A. Akbari, C. Zhang, A. Akbari, L. Zhao, Y. Chen, W. Chen, X. Zhang, G. Yuan, and Y. Wang. Flash-W AM: Modality-aware distillation for world action models, 2026. URL https://arxiv.org/abs/2606.05254 on PithPith's reviewarXiv
  5. [5] I. Apanasevich, M. Artemyev, R. Babakyan, P. Fedotova, D. Grankin, E. Kupryashin, A. Misailidi, D. Nerus, et al. Green-VLA: Staged vision-language-action model for generalist robots, 2026. URL https://arxiv.org/abs/2602.00919 on PitharXiv
  6. [6] M. Assran, A. Bardes, D. Fan, Q. Garrido, R. Howes, M. Komeili, M. Muckley, et al. V-JEPA 2: Self-supervised video models enable understanding, prediction and planning, 2025. URL https://arxiv.org/abs/2506.09985 on PithPith's reviewarXiv
  7. [7] H. Bi, H. Tan, S. Xie, Z. Wang, S. Huang, H. Liu, R. Zhao, Y. Feng, C. Xiang, Y. Rong, H. Zhao, H. Liu, Z. Su, L. Ma, H. Su, and J. Zhu. Motus: A Unified Latent Action World Model, 2025. URL https://arxiv.org/abs/2512.13030 on PithPith's reviewarXiv
  8. [8] H. Bi, Z. Zhou, Y. Tang, J. Pang, S. Huang, H. Liu, R. Wang, S. Huang, et al. Motus2: A self-evolving general world model for dexterous manipulation, 2026. URL https://arxiv.org/abs/2608.30237 on PithPith's reviewarXiv
  9. [9] BitRobot, Unitree, and Hugging Face. HIW-500: Humanoids in-the-wild dataset for robot learning. Dataset, 2026. URL https://bitrobot-foundation.github.io/humanoids-in-the-wild-500-hours on Pith
  10. [10] BitRobot Foundation. G1 WBT Dex1 building-children-table, 2026. URL https://huggingface.co/datasets/BitRobot/G1_WBT_ Dex1_Building-Children-Table . Dataset on Hugging Face on Pith
  11. [11] K. Black, N. Brown, D. Driess, A. Esmail, M. Equi, C. Finn, N. Fusai, L. Groom, et al. 𝜋0: A vision-language-action flow model for general robot control, 2024. URL https://arxiv.org/abs/2410.24164. 26 on PithPith's reviewarXiv
  12. [12] BWM Team. BWM: A low-cost high-fidelity world simulator for robot learning, 2026. URL https://arxiv.org/abs/2607.29302 on PithPith's reviewarXiv
  13. [13] R. Cadene, S. Alibert, A. Soare, Q. Gallouedec, A. Zouitine, S. Palma, P. Kooijmans, M. Aractingi, M. Shukor, D. Aubakirova, et al. LeRobot: State-of-the-art machine learning for real-world robotics in PyTorch. GitHub repository, 2024. URL https: //github.com/huggingface/lerobot on Pith
  14. [14] J. Cai, L. Ling, S. Chu, Z. Liu, J. Kang, Z. Liang, W. Xu, Y. Mao, et al. AHA-W AM: Asynchronous horizon-adaptive world-action modeling with observation-guided context routing, 2026. URL https://arxiv.org/abs/2606.09811 on PithPith's reviewarXiv
  15. [15] R. Cai, J. Guo, X. He, P. Jin, J. Li, B. Lin, F. Liu, W. Liu, et al. Xiaomi-Robotics-0: An open-sourced vision-language-action model with real-time execution, 2026. URL https://arxiv.org/abs/2602.12684 on PitharXiv
  16. [16] C.-L. Cheang, G. Chen, Y. Jing, T. Kong, H. Li, Y. Li, Y. Liu, H. Wu, et al. GR-2: A generative video-language-action model with web-scale knowledge for robot manipulation, 2024. URL https://arxiv.org/abs/2410.06158 on PithPith's reviewarXiv
  17. [17] K. Chen, Z. Liu, T. Zhang, Z. Guo, S. Xu, H. Lin, H. Zang, X. Li, B. Wei, J. Zhou, Q. Zhang, Z. Yu, G. Fan, T. Huang, Y. Wang, and C. Yu. 𝜋RL: Online RL fine-tuning for flow-based vision-language-action models, 2025. URL https://arxiv.org/abs/2510. 25889 on Pith
  18. [18] T. Chen, Z. Chen, B. Chen, Z. Cai, Y. Liu, Z. Li, Q. Liang, X. Lin, et al. RoboTwin 2.0: A scalable data generator and benchmark with strong domain randomization for robust bimanual robotic manipulation, 2025. URL https://arxiv.org/abs/2506.18088 on PithPith's reviewarXiv
  19. [19] T. Chen, Y. Chen, Z. Li, J. Tang, K. Su, H. Lu, W. Wan, B. Chen, et al. RoboDojo: A unified sim-and-real benchmark for comprehensive evaluation of generalist robot manipulation policies, 2026. URL https://arxiv.org/abs/2607.04434 on PithPith's reviewarXiv
  20. [20] Y. Chen, R. Chen, D. Huo, Y. Yang, D. Qi, H. Liu, T. Lin, S. Zeng, et al. ABot-PhysWorld: Interactive world foundation model for robotic manipulation with physics alignment, 2026. URL https://arxiv.org/abs/2603.23376 on PitharXiv
show 80 more
  1. [21] Y. Chen, P. Li, Y. Xu, Q. Ma, J. Yang, K. Wang, J. Yang, D. An, et al. FlowW AM: Optical flow as a unified action representation for world action models, 2026. URL https://arxiv.org/abs/2607.13017 on PithPith's reviewarXiv
  2. [22] R. Doshi, H. Walke, O. Mees, S. Dasari, and S. Levine. Scaling cross-embodied learning: One policy for manipulation, navigation, locomotion and aviation, 2024. URL https://arxiv.org/abs/2408.11812 on PithPith's reviewarXiv
  3. [23] Y. Du, M. Yang, B. Dai, H. Dai, O. Nachum, J. B. Tenenbaum, D. Schuurmans, and P. Abbeel. Learning universal policies via text-guided video generation, 2023. URL https://arxiv.org/abs/2302.00111 on PithPith's reviewarXiv
  4. [24] S. Fei, S. Wang, J. Shi, Z. Dai, J. Cai, P. Qian, L. Ji, X. He, S. Zhang, Z. Fei, J. Fu, J. Gong, and X. Qiu. LIBERO-Plus: In-depth robustness analysis of vision-language-action models, 2025. URL https://arxiv.org/abs/2510.13626 on PithPith's reviewarXiv
  5. [25] S. Gao, W. Liang, K. Zheng, A. Malik, S. Ye, S. Yu, W.-C. Tseng, Y. Dong, et al. DreamDojo: A generalist robot world model from large-scale human videos, 2026. URL https://arxiv.org/abs/2602.06949 on PithPith's reviewarXiv
  6. [26] GigaBrain Team, A. Ye, A. Sun, C. Jin, C. Cheng, C. Shi, D. Shang, D. Zhang, G. Huang, et al. GigaBrain-0.7: Scaling embodied foundation models to emergent capabilities with a three-system architecture, 2026. URL https://arxiv.org/abs/2608.15875 on PithPith's reviewarXiv
  7. [27] GigaWorld Team, A. Ye, B. Wang, C. Ni, G. Huang, G. Zhao, H. Li, J. Zhu, K. Li, et al. GigaWorld-0: World models as data engine to empower embodied AI, 2025. URL https://arxiv.org/abs/2511.19861 on PitharXiv
  8. [28] GigaWorld Team, A. Ma, B. Wang, B. Li, C. Ni, G. Li, G. Huang, G. Zhao, et al. GigaWorld-1: A roadmap to build world models for robot policy evaluation, 2026. URL https://arxiv.org/abs/2607.02642 on PithPith's reviewarXiv
  9. [29] GigaWorld Team, A. Ye, A. Ma, B. Wang, C. Ni, F. Ye, G. Huang, G. Li, G. Zhao, et al. GigaWorld-Policy-0.5: A faster and stronger W AM empowered by autoresearch, 2026. URL https://arxiv.org/abs/2607.13960 on PithPith's reviewarXiv
  10. [30] J. Guo, Q. Li, P. Li, Z. Chen, N. Sun, Y. Su, H. Wang, Y. Zhang, et al. Unified 4D world action modeling from video priors with asynchronous denoising, 2026. URL https://arxiv.org/abs/2604.26694 on PithPith's reviewarXiv
  11. [31] Y. Guo, L. X. Shi, J. Chen, and C. Finn. Ctrl-World: A controllable generative world model for robot manipulation, 2025. URL https://arxiv.org/abs/2510.10125 on PithPith's reviewarXiv
  12. [32] J. Ho and T. Salimans. Classifier-free diffusion guidance, 2022. URL https://arxiv.org/abs/2207.12598 on PithPith's reviewarXiv
  13. [33] Y. Hu, Y. Guo, P. Wang, X. Chen, Y.-J. Wang, J. Zhang, K. Sreenath, C. Lu, et al. Video prediction policy: A generalist robot policy with predictive visual representations, 2024. URL https://arxiv.org/abs/2412.14803 on PithPith's reviewarXiv
  14. [34] J. Jang, S. Ye, Z. Lin, J. Xiang, J. Bjorck, Y. Fang, F. Hu, S. Huang, et al. DreamGen: Unlocking generalization in robot learning through video world models, 2025. URL https://arxiv.org/abs/2505.12705 on PithPith's reviewarXiv
  15. [35] T. Jiang, T. Yuan, Y. Liu, C. Lu, J. Cui, X. Liu, S. Cheng, J. Gao, et al. Galaxea open-world dataset and G0 dual-system VLA model, 2025. URL https://arxiv.org/abs/2509.00576 on PithPith's reviewarXiv
  16. [36] Y. Jiang, Y. Guo, H. Yang, G. Lei, N. Chen, Y. Zhang, S. Yan, B. Lin, F. Gao, and B. Qi. CKT-W AM: Parameter-efficient context knowledge transfer between world action models, 2026. URL https://arxiv.org/abs/2605.06247 on PithPith's reviewarXiv
  17. [37] J. Ke, Q. Wang, Y. Wang, P. Milanfar, and F. Yang. MUSIQ: Multi-scale image quality transformer, 2021. URL https: //arxiv.org/abs/2108.05997 on PithPith's reviewarXiv
  18. [38] M. J. Kim, C. Finn, and P. Liang. Fine-tuning vision-language-action models: Optimizing speed and success, 2025. URL https://arxiv.org/abs/2502.19645 on PithPith's reviewarXiv
  19. [39] M. J. Kim, Y. Gao, T.-Y. Lin, Y.-C. Lin, Y. Ge, G. Lam, P. Liang, S. Song, et al. Cosmos Policy: Fine-tuning video models for visuomotor control and planning, 2026. URL https://arxiv.org/abs/2601.16163 on PithPith's reviewarXiv
  20. [40] C. Li, R. Zhang, J. Wong, C. Gokmen, S. Srivastava, R. Martín-Martín, C. Wang, G. Levine, et al. BEHA VIOR-1K: A human- centered, embodied ai benchmark with 1,000 everyday activities and realistic simulation, 2024. URL https://arxiv.org/abs/ 2403.09227 on PithPith's reviewarXiv
  21. [41] F. Li, W. Song, H. Zhao, J. Wang, P. Ding, D. Wang, L. Zeng, and H. Li. Spatial forcing: Implicit spatial representation alignment for vision-language-action model, 2025. URL https://arxiv.org/abs/2510.12276 on PitharXiv
  22. [42] L. Li, Q. Zhang, Y. Luo, S. Yang, R. Wang, F. Han, M. Yu, Z. Gao, et al. Causal world modeling for robot control, 2026. URL https://arxiv.org/abs/2601.21998 on PithPith's reviewarXiv
  23. [43] S. Li, Y. Gao, D. Sadigh, and S. Song. Unified video action model, 2025. URL https://arxiv.org/abs/2503.00200 on PithPith's reviewarXiv
  24. [44] S. Li, V. Yao, C. Yang, T. Qu, R. Cheng, R. Yu, H. Lu, N. Von, et al. W ALL-WM: Carving world action modeling at the event joints, 2026. URL https://arxiv.org/abs/2606.01955 on PithPith's reviewarXiv
  25. [45] X. Li, P. Li, L. Qian, M. Liu, D. Wang, J. Liu, B. Kang, X. Ma, et al. What matters in building vision-language-action models for generalist robots, 2024. URL https://arxiv.org/abs/2412.14058. 27 on PithPith's reviewarXiv
  26. [46] X. Li, J. Guo, Q. Li, L. Qian, H. Lai, Y. Wang, H. Yan, J. Cao, et al. Xiaomi-Robotics-U0: Unified embodied synthesis with world foundation model, 2026. URL https://arxiv.org/abs/2607.11643 on PithPith's reviewarXiv
  27. [47] Y. Liao, P. Zhou, S. Huang, D. Yang, S. Chen, Y. Jiang, Y. Hu, J. Cai, S. Liu, J. Luo, L. Chen, S. Yan, M. Yao, and G. Ren. Genie Envisioner: A unified world foundation platform for robotic manipulation, 2025. URL https://arxiv.org/abs/2508.05635v3 on PithPith's reviewarXiv
  28. [48] B. Liu, Y. Zhu, C. Gao, Y. Feng, Q. Liu, Y. Zhu, and P. Stone. LIBERO: Benchmarking knowledge transfer for lifelong robot learning. In Advances in Neural Information Processing Systems (Datasets and Benchmarks Track) , 2023 on Pith
  29. [49] S. Liu, L. Wu, B. Li, H. Tan, H. Chen, Z. Wang, K. Xu, H. Su, et al. RDT-1B: a diffusion foundation model for bimanual manipulation, 2024. URL https://arxiv.org/abs/2410.07864 on PithPith's reviewarXiv
  30. [50] X. Liu, C. Gong, and Q. Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow, 2022. URL https://arxiv.org/abs/2209.03003 on PithPith's reviewarXiv
  31. [51] Y. Liu, P. Ding, T. Jiang, X. Wang, M. Lin, W. Song, H. Zhang, Z. Zhuang, H. Zhao, W. Zhao, S. Huang, J. Shi, and D. Wang. MMaDA-VLA: Large diffusion vision-language-action model with unified multi-modal instruction and generation, 2026. URL https://arxiv.org/abs/2603.25406 on PithPith's reviewarXiv
  32. [52] Y. Liu, P. Sun, S. Li, Y. Xie, L. Zhang, X. Chao, S. Dong, F. Chen, X.-P. Zhang, and W. Ding. OA-W AM: Object-addressable world action model for robust robot manipulation, 2026. URL https://arxiv.org/abs/2605.06481 on PithPith's reviewarXiv
  33. [53] I. Loshchilov and F. Hutter. Decoupled weight decay regularization, 2017. URL https://arxiv.org/abs/1711.05101 on PithPith's reviewarXiv
  34. [54] G. Y. Luo, G. M. Favero, Z. H. Luo, A. Jolicoeur-Martineau, and C. Pal. Beyond FVD: Enhanced evaluation metrics for video generation quality, 2024. URL https://arxiv.org/abs/2410.05203 on PithPith's reviewarXiv
  35. [55] O. Mees, L. Hermann, E. Rosete-Beas, and W. Burgard. CAL VIN: A benchmark for language-conditioned policy learning for long-horizon robot manipulation tasks, 2021. URL https://arxiv.org/abs/2112.03227 on PithPith's reviewarXiv
  36. [56] Motubrain Team, C. Xiang, F. Bao, H. Liu, H. Tan, H. Bi, J. Li, J. Liu, J. Pang, et al. Motubrain: An advanced world action model for robot control, 2026. URL https://arxiv.org/abs/2604.27792 on PithPith's reviewarXiv
  37. [57] NVIDIA, A. Ali, J. Bai, M. Bala, Y. Balaji, A. Blakeman, T. Cai, J. Cao, T. Cao, et al. World simulation with video foundation models for physical ai, 2025. URL https://arxiv.org/abs/2511.00062 on PithPith's reviewarXiv
  38. [58] NVIDIA, Aditi, N. Agarwal, A. Ali, J. Allen, M. Antolini, A. Aubame, A. Azzolini, J. Bai, et al. Cosmos 3: Omnimodal world models for physical AI, 2026. URL https://arxiv.org/abs/2606.02800 on PithPith's reviewarXiv
  39. [59] Octo Model Team, D. Ghosh, H. Walke, K. Pertsch, K. Black, O. Mees, S. Dasari, J. Hejna, T. Kreiman, et al. Octo: An open-source generalist robot policy, 2024. URL https://arxiv.org/abs/2405.12213 on PithPith's reviewarXiv
  40. [60] Open X-Embodiment Collaboration, A. O’Neill, A. Rehman, A. Gupta, A. Maddukuri, A. Gupta, A. Padalkar, A. Lee, A. Pooley, et al. Open X-Embodiment: Robotic learning datasets and RT-X models, 2023. URL https://arxiv.org/abs/2310.08864 on PithPith's reviewarXiv
  41. [61] Physical Intelligence. OpenPI: Open-source robotics models and fine-tuning. Official software repository, 2025. URL https: //github.com/Physical-Intelligence/openpi on Pith
  42. [62] Physical Intelligence, K. Black, N. Brown, J. Darpinian, K. Dhabalia, D. Driess, A. Esmail, M. Equi, C. Finn, et al. 𝜋0.5: a vision-language-action model with open-world generalization, 2025. URL https://arxiv.org/abs/2504.16054 on PithPith's reviewarXiv
  43. [63] B. Qiu, L. Chen, Y. Liao, N. Wang, L. Wang, J. Luo, W. Zhao, S. Chen, et al. GE-Sim 2.0: A roadmap towards comprehensive closed-loop video world simulators for robotic manipulation, 2026. URL https://arxiv.org/abs/2605.27491 on PithPith's reviewarXiv
  44. [64] Qwen Team. Qwen3.8-Max: A new bar for coding and cowork. Blog post, August 2026. URL https://qwen.ai/blog?id=qwen3.8 . Model card: https://huggingface.co/Qwen/Qwen3.8-27B on Pith
  45. [65] M. Reuss, H. Zhou, M. Rühle, Ömer Erdinç Yağmurlu, F. Otto, and R. Lioutikov. FLOWER: Democratizing generalist robot policies with efficient vision-language-action flow policies, 2025. URL https://arxiv.org/abs/2509.04996 on PithPith's reviewarXiv
  46. [66] Robbyant. LingBot-V A repository, 2026. URL https://github.com/Robbyant/lingbot-va on Pith
  47. [67] RoboDojo Team. RoboDojo leaderboard. Project website, 2026. URL https://robodojo-benchmark.com/leaderboard on Pith
  48. [68] RoboTwin Team. RoboTwin 2.0 leaderboard. Project website, 2026. URL https://robotwin-platform.github.io/leaderboard on Pith
  49. [69] C. Schuhmann, R. Beaumont, R. Vencu, C. Gordon, R. Wightman, M. Cherti, T. Coombes, A. Katta, C. Mullis, M. Wortsman, et al. LAION-5B: An open large-scale dataset for training next generation image-text models, 2022. URL https://arxiv.org/ abs/2210.08402 on PithPith's reviewarXiv
  50. [70] Y. Shang, Z. Li, Y. Ma, W. Su, X. Jin, Z. Wang, L. Jin, X. Zhang, et al. WorldArena: A unified benchmark for evaluating perception and functional utility of embodied world models, 2026. URL https://arxiv.org/abs/2602.08971 on PitharXiv
  51. [71] Y. Shang, Y. Tang, Y. Ma, Z. Li, L. Jin, W. Su, X. Jin, Z. Wang, et al. WorldArena 2.0: Extending embodied world model benchmarking on modality, functionality and platform, 2026. URL https://arxiv.org/abs/2605.17912 on PithPith's reviewarXiv
  52. [72] Y. Song and P. Dhariwal. Improved techniques for training consistency models, 2023. URL https://arxiv.org/abs/2310.14189 on PithPith's reviewarXiv
  53. [73] StarVLA Community. StarVLA: A Lego-like codebase for vision-language-action model developing, 2026. URL https://arxiv. org/abs/2604.05014 on PithPith's reviewarXiv
  54. [74] H. Su, Z. Liu, X. Jin, H. Dou, C. Hu, B. Li, Z. Liu, R. Xu, J. Fang, X. Zhang, Z. Yang, X. Yang, C. Gao, J. Yan, Y. Li, and W. Wu. WorldScape Policy 2.0: Empowering steerable world action modeling with reasoning-augmented memory and in-context learning, 2026. URL https://arxiv... on PithPith's reviewarXiv
  55. [75] T. Su, J. Zhu, T. Wang, Y. He, Z. Huang, J. Zhang, C. Ma, H. Wang, et al. DeMaVLA: A vision-language-action foundation model for generalizable deformable manipulation, 2026. URL https://arxiv.org/abs/2605.31286 on PithPith's reviewarXiv
  56. [76] Y. Tian, S. Yang, J. Zeng, P. Wang, D. Lin, H. Dong, and J. Pang. Predictive inverse dynamics models are scalable learners for robotic manipulation, 2024. URL https://arxiv.org/abs/2412.15109 on PithPith's reviewarXiv
  57. [77] Y. Tian, Y. Yang, Y. Xie, Z. Cai, X. Shi, N. Gao, H. Liu, X. Jiang, et al. InternData-A1: Pioneering high-fidelity synthetic data for pre-training generalist policy, 2025. URL https://arxiv.org/abs/2511.16651 on PitharXiv
  58. [78] Unitree Robotics. Unitree open robot datasets, 2025. URL https://huggingface.co/unitreerobotics. Dataset on Hugging Face on Pith
  59. [79] T. Wan, A. Wang, B. Ai, B. Wen, C. Mao, C.-W. Xie, D. Chen, F. Yu, et al. Wan: Open and advanced large-scale video generative models, 2025. URL https://arxiv.org/abs/2503.20314 on PithPith's reviewarXiv
  60. [80] L. Wang, X. Chen, J. Zhao, and K. He. Scaling proprioceptive-visual learning with heterogeneous pre-trained transformers, 2024. URL https://arxiv.org/abs/2409.20537. 28 on PithPith's reviewarXiv
  61. [81] Y. Wang, X. Li, W. Wang, J. Zhang, Y. Li, Y. Chen, X. Wang, and Z. Zhang. Unified vision-language-action model, 2025. URL https://arxiv.org/abs/2506.19850 on PithPith's reviewarXiv
  62. [82] Z. Wang, C. Liu, Y. Xiang, R. Zhang, Q. Hao, H. Lu, H. Chen, Z. Feng, et al. The great march 100: 100 detail-oriented tasks for evaluating embodied ai agents, 2026. URL https://arxiv.org/abs/2601.11421 on PitharXiv
  63. [83] WorldArena Team. WorldArena 2.0: Motion quality evaluation update. Project website, 2026. URL https://v2.world-arena. ai/WorldArena-2.0/motion-quality-evaluation-update.html on Pith
  64. [84] WorldArena Team. WorldArena leaderboard. Hugging Face Space, 2026. URL https://huggingface.co/spaces/WorldArena/ WorldArena on Pith
  65. [85] K. Wu, C. Hou, J. Liu, Z. Che, X. Ju, Z. Yang, M. Li, Y. Zhao, et al. RoboMIND: Benchmark on multi-embodiment intelligence normative data for robot manipulation, 2024. URL https://arxiv.org/abs/2412.13877 on PithPith's reviewarXiv
  66. [86] S. Wu, X. Liu, S. Xie, P. Wang, X. Li, B. Yang, Z. Li, K. Zhu, et al. RoboCOIN: An open-sourced bimanual robotic data collection for integrated manipulation, 2025. URL https://arxiv.org/abs/2511.17441 on PithPith's reviewarXiv
  67. [87] G. Yang, Z. Tu, Y. Yang, S. Mao, J. Dong, T. Chen, J. Peng, J. Xiong, et al. EventVLA: Event-driven visual evidence memory for long-horizon vision-language-action policies, 2026. URL https://arxiv.org/abs/2606.20092 on PithPith's reviewarXiv
  68. [88] L. Yang, W. Song, X. Wang, P. Sheng, Z. Fang, Z. Zhou, J. He, H. Yan, et al. 4D-W AM: Infusing spatiotemporal awareness into world action models through trajectory fields, 2026. URL https://arxiv.org/abs/2608.08023 on PithPith's reviewarXiv
  69. [89] S. Yang, Y. Du, K. Ghasemipour, J. Tompson, L. Kaelbling, D. Schuurmans, and P. Abbeel. Learning interactive real-world simulators, 2023. URL https://arxiv.org/abs/2310.06114 on PithPith's reviewarXiv
  70. [90] Y. Yang, S. Zeng, T. Lin, X. Chang, D. Qi, J. Xiao, H. Liu, R. Chen, et al. ABot-M0: VLA foundation model for robotic manipulation with action manifold learning, 2026. URL https://arxiv.org/abs/2602.11236 on PithPith's reviewarXiv
  71. [91] Z. Yang, J. Teng, W. Zheng, M. Ding, S. Huang, J. Xu, Y. Yang, W. Hong, et al. CogVideoX: Text-to-video diffusion models with an expert transformer, 2024. URL https://arxiv.org/abs/2408.06072 on PithPith's reviewarXiv
  72. [92] S. Ye, Y. Ge, K. Zheng, S. Gao, S. Yu, G. Kurian, S. Indupuru, Y. L. Tan, et al. World action models are zero-shot policies, on Pith
  73. [93] C. Yin, D. Huang, D. Yang, J. Wang, N. Zhao, C. Xu, W. Sun, L. Hou, et al. Genie sim 3.0 : A high-fidelity comprehensive simulation platform for humanoid robot, 2026. URL https://arxiv.org/abs/2601.02078 on PithPith's reviewarXiv
  74. [94] H. Yuan, Z. Liang, A. Chen, Y. Wang, H. Li, P. Lin, Y. Huang, Z. Lei, et al. Qwen-robotmanip technical report: Alignment unlocks scale for robotic manipulation foundation models, 2026. URL https://arxiv.org/abs/2606.17846 on PithPith's reviewarXiv
  75. [95] T. Yuan, Z. Dong, Y. Liu, and H. Zhao. Fast-W AM: Do world action models need test-time future imagination?, 2026. URL https://arxiv.org/abs/2603.16666 on PithPith's reviewarXiv
  76. [96] M. Zhang, R. Sun, X. Wang, C. Zhang, S. Li, M. Cao, H. Lu, E. Chen, et al. W ALL-SS: Scaling long-horizon world models via next-scale autoregression, 2026. URL https://arxiv.org/abs/2608.26239 on PithPith's reviewarXiv
  77. [97] Q. Zhang, L. Li, L. Zhang, S. Yang, Y. Luo, S. Li, R. Wang, J. Wang, et al. Native video-action pretraining for generalizable robot control, 2026. URL https://arxiv.org/abs/2607.08639 on PithPith's reviewarXiv
  78. [98] W. Zhang, H. Liu, Z. Qi, Y. Wang, X. Yu, J. Zhang, R. Dong, J. He, F. Lu, H. Wang, Z. Zhang, L. Yi, W. Zeng, and X. Jin. DreamVLA: A vision-language-action model dreamed with comprehensive world knowledge, 2025. URL https://arxiv.org/abs/ 2507.04447 on PithPith's reviewarXiv
  79. [99] X. Zhang, X. Liu, Q. Tang, J. Wang, R. Wang, Z. Cheng, B. Lin, H. Wen, M. Li, K. Zhan, and Y. Liu. ME-Dex 1.0: Bringing Heterogeneous Tactile Sensing into World Action Modeling, 2026. URL https://arxiv.org/abs/2609.21449 on PithPith's reviewarXiv
  80. [100] W. Zhao, H. Jiang, X. Shi, L. Liu, F. Huang, Z. Su, W. Sui, and X. Wang. Faster-W AM: Efficient inference-time future conditioning for robust world action models, 2026. URL https://arxiv.org/abs/2608.04404 on PithPith's reviewarXiv

The first 100 of 107 references are listed here. All of them are in this paper's citation record.

Challenges and discussion

Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.

No comments yet.

Sign in to comment