Pith. sign in

REVIEW 3 major objections 5 minor 26 references

Agentic Real2Sim converts real robot-demonstration videos into physics-simulated digital twins automatically.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 13:08 UTC pith:7JLSEWKY

load-bearing objection Promising agentic real2sim architecture with a clear tool/agent split, but the VLM-judged replay metric leaves the physical-fidelity claims unproven. the 3 major comments →

arxiv 2607.19190 v3 pith:7JLSEWKY submitted 2026-07-21 cs.RO cs.AI

Agentic Real2Sim: Physics-based World Modeling with Vision-Language Agents

Guanxiong Chen , Qianjun Xia , Jiawei Peng , Heng Zhang , Bole Ma , Justin Qian , Ziyi Jiao , Bingyang Zhou
show 16 more authors
Luoxin Ye Kaifeng Zhang Kunyi Wang Weijia Zeng Yunuo Chen Pengzhi Yang Ziqiu Zeng Siyuan Luo Huamin Wang Chao Liu Alan Yuille Fan Shi Changxi Zheng Yunzhu Li Chenfanfu Jiang Peter Yichen Chen
This is my paper
classification cs.RO cs.AI
keywords real-to-simvision-language agentsrobot manipulationdigital twinphysics simulationagentic pipelinedeformable objectshumanoid motion
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper's central claim is that real-world recordings of robot-object interaction—not just static scenes—can be automatically converted into runnable physical simulations, called episode twins, that preserve the actors, objects, trajectories, and physical parameters of the original episode. The framework achieves this with a pipeline of vision-language agents that orchestrate deterministic perception tools (segmentation, depth reconstruction, pose tracking) and iteratively refine the simulated scene by replaying it in a physics engine. Evaluated on 100 real manipulation episodes, the framework succeeds in roughly half of cases, and an open 31-billion-parameter vision-language model matches proprietary backends in conversion success while cutting model cost by up to 31×. The same 'episode contract' generalizes to deformable objects and humanoid motion, suggesting a unified path toward scalable real-to-sim conversion without manual scene authoring.

Core claim

This paper proposes that the bottleneck in real-to-sim conversion is not missing perception models but the glue between them. It introduces Agentic Real2Sim, a framework whose four agents—visual processing, physical-prior inference, scene preparation, and simulator-in-the-loop grasp optimization—convert a recorded interaction into a simulatable 'episode twin' represented as a structured tuple (observations, actors, geometry, states, parameters, simulator, metrics). The framework is evaluated on 100 real manipulation episodes: an open 31B vision-language model produces 48 successful, 8 partial, and 44 failed replay outcomes, with other backends scoring 37–45 successes, and the open model cost

What carries the argument

The central mechanism is the 'episode twin' contract—a structured representation that standardizes what a conversion must preserve across domains, together with a VLM-driven agentic loop that makes bounded, schema-constrained decisions (object discovery, keyframe selection, mask acceptance, refinement choices) while delegating geometry, physics, and rendering to deterministic tools. This separation lets the pipeline transfer across vision-language backends without per-model tuning.

Load-bearing premise

The replay-success metric assumes that a vision-language judge's score of 8 or higher out of 10 for real-versus-simulated keyframe similarity is a valid measure of genuine physical alignment between the simulation and the real episode.

What would settle it

Take a set of converted episode twins that pass the VLM replay-success threshold, then directly compare the simulated object trajectories and contact events against tracked real-world poses; if a substantial fraction of 'successful' twins show large quantitative divergence (for example, object position errors exceeding several centimeters) despite high VLM scores, the metric's claim to physical alignment is refuted.

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

If this is right

  • If conversion can be automated at a cost of about $2.62 per 100 episodes with an open model, large-batch simulation assets for policy learning become practical.
  • The episode contract provides a common interface across rigid, deformable, and humanoid domains, reducing the need for separate real-to-sim pipelines.
  • Observed replay success near 50% across backends indicates the remaining bottleneck is upstream perception and simulation fidelity, not vision-language reasoning power.
  • Simulator-in-the-loop refinement (grasp optimization) shows that a physics engine can correct imperfect vision estimates during conversion.

Where Pith is reading between the lines

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

  • A direct extension is to replace the VLM judge with quantitative physical metrics—such as object-pose error or contact forces—to validate whether 'replay success' truly corresponds to physical alignment.
  • The same agentic orchestration pattern could transfer to other domains that combine perception with simulation, such as biomechanics or autonomous driving scene reconstruction.
  • The cost-efficiency result suggests that when tool boundaries are cleanly separated, small open-weight models can substitute for frontier models, a lesson likely relevant to other agentic pipelines beyond real-to-sim.
  • If replay success is driven by visual plausibility rather than dynamics, the twin may be insufficient for contact-rich policy learning; a targeted study comparing policy performance in real versus simulated episodes would clarify this.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper presents Agentic Real2Sim, a VLM-orchestrated pipeline that converts real-world recordings of robot-object interactions into simulatable MuJoCo episode twins. The system decomposes conversion into four stages—visual processing, physical-prior inference, scene preparation, and simulator-in-the-loop grasp optimization—organized under a shared 'episode contract.' It is evaluated on DROID-100 rigid manipulation episodes using a VLM-judged replay-success metric, across four VLM backends, with reported success counts from 37/100 to 48/100 and model costs ranging from $2.62 to $82.30. The authors also demonstrate qualitative deformable (PhysTwin-style) and humanoid (BFM-Zero-style) adapters. The central claims are that the framework preserves observations, geometries, robot interactions, and object states; that an open 31B VLM achieves comparable results to proprietary backends at up to 31.4× lower cost; and that the same contract generalizes across domains.

Significance. If the quantitative claims are substantiated, Agentic Real2Sim would be a meaningful step toward automated real-to-sim conversion: it provides a modular, backend-agnostic pipeline with explicit artifact contracts and bounded agentic decision-making, and it suggests that open-weight models can serve as practical orchestrators. The 31.4× cost reduction claim is striking and relevant. However, the evidence is currently insufficient to support the 'physical alignment' and 'preserving object states' claims, because the success metric is a VLM visual-plausibility score with no ground-truth validation. The contribution is therefore conditional; the framework design is promising, but the evaluation must be strengthened before the claims can be accepted.

major comments (3)
  1. [§4.1 (replay-success metric)] The central quantitative evaluation rests entirely on a VLM-judged metric that scores real-vs-sim keyframes on target-object identity, final location, action similarity, and end-gripper location. Starting-pose drift is explicitly 'recorded as context rather than directly subtracted,' and §3.1's grasp sweep deliberately shifts object placement. A successful episode can therefore have a different initial object state from the real one. This is a visual-plausibility score, not a measure of physical fidelity: it does not verify trajectories, contacts, forces, or inferred physical parameters. As written, the abstract's claim that the framework 'preserves ... object states' and the conclusion's claim that 'physical parameters' are preserved are not supported by the reported 48/100 success count. The authors should validate the metric against human judgments or physical ground-truth (e.g., traj
  2. [§4.2 and §4.3] Each VLM backend is run exactly once on DROID-100; the success counts (48, 45, 43, 37) are not accompanied by variance, confidence intervals, or statistical tests. The differences among backends are within what could be run-to-run noise, and the headline 31.4× cost ratio is a point estimate from a single run. The claim that 'an open 31B VLM achieves comparable observed replay-success outcomes' is therefore not statistically established. Repeating the pipeline with multiple seeds or reporting bootstrap intervals would be necessary.
  3. [§4.4 and §5] The deformable and humanoid results are qualitative only, and the limitations paragraph in §5 states that future work will 'extend automated conversion to deformable episodes and systematically evaluate agentic components.' This contradicts the abstract's statement that the framework is 'evaluated' across these domains. The generalization claim is thus an over-claim; at minimum, the abstract and §1 should be revised to say these are preliminary stress tests, or quantitative deformable/humanoid metrics should be added.
minor comments (5)
  1. [§4.3] The identities of the three judge VLMs are not specified. If a judge backend coincides with a conversion backend, the evaluation would be less convincing; please state the judge models explicitly.
  2. [Fig. 3] The left panel could include error bars or multiple-run markers; the right panel's log scale makes the 31.4× difference visually large, but the interpretation should mention the single-run nature.
  3. [Eq. (1)] The episode-twin tuple includes symbols (S_{1:T}, Θ, B, M) that are partly explained only later; consider defining them in a table or in the text immediately after the equation.
  4. [Related work] No direct quantitative comparison with prior real2sim pipelines (e.g., Scalable Real2Sim [15], TwinAligner [5]) is provided. Adding a comparison on a common benchmark would help contextualize the 'first step toward scalable conversion' claim.
  5. [Fig. 4 caption] The caption refers to 'Fig. 4 (a)' and '(b)' but the figure panels are labeled A and B; unify the notation.

Circularity Check

0 steps flagged

No circularity: quantitative claims rest on an external VLM-judge metric and are not forced by fitted parameters; self-citations are background only.

full rationale

I walked the claimed derivation chain: real recording -> visual processing -> physical-prior inference -> scene preparation -> simulator-in-the-loop grasp optimization -> MuJoCo replay -> replay-success metric. The central quantitative quantity, replay success, is defined in Sec. 4.1 as a VLM-judge measurement: "Three judges, each with a distinct VLM backend, then independently compare the episode’s real keyframes with each candidate’s simulated keyframes... scores of 8 or higher count as pass." This is an evaluation protocol, not an equation that reduces the output to the input. The judges are not used to fit or select geometry, physical parameters, or placements; candidate selection is deterministic ("eligible only if the probe marks it as grasped... lift and displacement statistics are finite and within broad sanity bounds"), and no fitted parameter is renamed as a prediction. Eq. (1) includes metrics/traces in the episode-twin tuple, but those are recorded after conversion rather than optimized by the pipeline, so the metric is not self-fulfilling by construction. The reported success counts and cost comparison are measurements under that metric, not forced values. The self-citations (PhysTwin [9], EMPM [4], LychSim [14]) are background/setting citations, not load-bearing; there is no imported uniqueness theorem or ansatz smuggled through self-citation. I do flag a validity risk, not circularity: the metric scores visual keyframe similarity, explicitly treats "Starting-pose drift... as context rather than directly subtracted," and the grasp sweep deliberately shifts object placement, so the abstract's "preserves observations, geometries, robot interactions, and object states" is stronger than what the metric verifies. That is an evidence-strength/measurement-validity problem, not a reduction of the claim to its inputs. Hence no circular step qualifies under the hard rules; score 0.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

The paper introduces no new physical entities. Its central claims rest on a set of unmeasured free parameters (object masses, placement shifts, humanoid gains) and on the validity of VLM-based evaluation as a proxy for physical correctness. The absence of any independent physical measurement is the main epistemic gap.

free parameters (3)
  • object mass/material class = not reported
    Assigned by the VLM physical-prior agent (Sec 3.2, Table 1). No calibration against real measurements; directly affects simulation dynamics.
  • object placement shift (grasp sweep) = not reported
    Optimized via deterministic sweep in Sec 3.2 to achieve successful grasp; not constrained by real episode observations.
  • humanoid joint gains = not reported
    Tuned in closed-loop simulation to match reference motion (Sec 4.4, Fig 4B).
axioms (4)
  • domain assumption VLM judges' scores are a valid proxy for physical fidelity
    Sec 4.1 defines replay success via VLM judges; no ground-truth physical metrics provided.
  • domain assumption DROID-100 sample is representative of rigid manipulation episodes
    Sec 4.2: 'randomly selected set of 100 manipulation episodes sampled to span objects, camera viewpoints, occlusion patterns, and manipulation verbs'; no statistical justification.
  • domain assumption Perception tools (SAM 3, SAM 3D, FoundationStereo, FoundationPose) provide sufficiently accurate geometry and pose for simulation
    Used as fixed deterministic tools (Sec 3.2); errors propagate to episode twins.
  • domain assumption MuJoCo can faithfully represent the physical interactions
    Chosen as simulator backend (Sec 3.2).

pith-pipeline@v1.3.0-alltime-deepseek · 8980 in / 8018 out tokens · 80890 ms · 2026-08-01T13:08:32.548275+00:00 · methodology

0 comments
read the original abstract

Real-to-sim conversion for robotic interaction with objects remains labor-intensive because it requires more than visual reconstruction: a streamlined real2sim process must recover scene geometries and object states, infer physical parameters, and assemble actors, objects, cameras, poses, and trajectories into a runnable physical simulation. Today this process still depends on manual tuning of visual foundation models, mesh cleanup, coordinate-frame alignment, and brittle workflow glue across visual perception tools and simulators. We introduce \textit{Agentic Real2Sim}, a framework for generalized physical world modeling with vision-language agents, converting a real-world recording of object-robot interaction into a simulatable episodic twin which preserves observations, geometries, robot interactions, and object states. We evaluate Agentic Real2Sim on rigid-object manipulation, deformable-object interaction, and humanoid motion scenes, spanning domains that are usually handled by separate Real2Sim pipelines, marking a first step toward scalable conversion. The framework's agentic decisions can be driven by an open-weight VLM backend at a small fraction of the cost of frontier models, while attaining comparable conversion success rate. We aim to use the resulting real-world-aligned twins for downstream robotics tasks, specifically policy learning and evaluation. The project site is available at https://agentic-real2sim.github.io/.

discussion (0)

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

Reference graph

Works this paper leans on

26 extracted references · 8 linked inside Pith

  1. [1]

    Physx-omni: Unified simulation-ready physical 3d generation for rigid, deformable, and articulated objects.arXiv preprint arXiv:2605.21572, 2026

    Ziang Cao, Yinghao Liu, Haitian Li, Runmao Yao, Fangzhou Hong, Zhaoxi Chen, Liang Pan, and Ziwei Liu. Physx-omni: Unified simulation-ready physical 3d generation for rigid, deformable, and articulated objects.arXiv preprint arXiv:2605.21572, 2026

    Pith/arXiv arXiv 2026

  2. [2]

    Sam 3: Segment anything with concepts.arXiv preprint arXiv:2511.16719, 2025

    Nicolas Carion, Laura Gustafson, Yuan-Ting Hu, Shoubhik Debnath, Ronghang Hu, Didac Suris, Chaitanya Ryali, Kalyan Vasudev Alwala, Haitham Khedr, Andrew Huang, et al. Sam 3: Segment anything with concepts.arXiv preprint arXiv:2511.16719, 2025

    Pith/arXiv arXiv 2025

  3. [3]

    Sam 3d: 3dfy anything in images.arXiv preprint arXiv:2511.16624, 2025

    Xingyu Chen, Fu-Jen Chu, Pierre Gleize, Kevin J Liang, Alexander Sax, Hao Tang, Weiyao Wang, Michelle Guo, Thibaut Hardin, Xiang Li, et al. Sam 3d: 3dfy anything in images.arXiv preprint arXiv:2511.16624, 2025

    Pith/arXiv arXiv 2025

  4. [4]

    Empm: Embodied mpm for modeling and simulation of deformable objects.IEEE Robotics and Automation Letters, 11(4):4179–4186, 2026

    Yunuo Chen, Yafei Hu, Lingfeng Sun, Tushar Kusnur, Laura Herlant, and Chenfanfu Jiang. Empm: Embodied mpm for modeling and simulation of deformable objects.IEEE Robotics and Automation Letters, 11(4):4179–4186, 2026. 10 Agentic Real2Sim: Generalized Physical World Modeling with Vision-Language Agents

    2026

  5. [5]

    Twinaligner: Visual-dynamic alignment empowers physics-aware real2sim2real for robotic manipulation.arXiv preprint arXiv:2512.19390, 2025

    Hongwei Fan, Hang Dai, Jiyao Zhang, Jinzhou Li, Qiyang Yan, Yujie Zhao, Mingju Gao, Jinghang Wu, Hao Tang, and Hao Dong. Twinaligner: Visual-dynamic alignment empowers physics-aware real2sim2real for robotic manipulation.arXiv preprint arXiv:2512.19390, 2025

    arXiv 2025

  6. [6]

    Skillmimicgen: Automated demonstration generation for efficient skill learning and deployment

    Caelan Garrett, Ajay Mandlekar, Bowen Wen, and Dieter Fox. Skillmimicgen: Automated demonstration generation for efficient skill learning and deployment. In8th Annual Conference on Robot Learning, 2024

    2024

  7. [7]

    Harvey, Mike Yurick, Derek Nowrouzezahrai, and Christopher Pal

    Félix G. Harvey, Mike Yurick, Derek Nowrouzezahrai, and Christopher Pal. Robust motion in-betweening. ACM Transactions on Graphics, 39(4), 2020

    2020

  8. [8]

    Pointworld: Scaling 3d world models for in-the-wild robotic manipulation.arXiv preprint arXiv:2601.03782, 2026

    Wenlong Huang, Yu-Wei Chao, Arsalan Mousavian, Ming-Yu Liu, Dieter Fox, Kaichun Mo, and Li Fei-Fei. Pointworld: Scaling 3d world models for in-the-wild robotic manipulation.arXiv preprint arXiv:2601.03782, 2026

    arXiv 2026

  9. [9]

    Phystwin: Physics-informed reconstruction and simulation of deformable objects from videos

    Hanxiao Jiang, Hao-Yu Hsu, Kaifeng Zhang, Hsin-Ni Yu, Shenlong Wang, and Yunzhu Li. Phystwin: Physics-informed reconstruction and simulation of deformable objects from videos. In2025 IEEE/CVF International Conference on Computer Vision (ICCV), page 7219–7230. IEEE, 2025

    2025

  10. [10]

    Simworld studio: Automatic environment generation with evolving coding agent for embodied agent learning.arXiv preprint arXiv:2605.09423, 2026

    Haoqiang Kang, Xiaokang Ye, Yuhan Liu, Siddhant Hitesh Mantri, Lingjun Mao, James Fleming, Drishti Regmi, and Lianhui Qin. Simworld studio: Automatic environment generation with evolving coding agent for embodied agent learning.arXiv preprint arXiv:2605.09423, 2026

    Pith/arXiv arXiv 2026

  11. [11]

    Gen2sim: Scaling up robot learning in simulation with generative models

    Pushkal Katara, Zhou Xian, and Katerina Fragkiadaki. Gen2sim: Scaling up robot learning in simulation with generative models. In2024 IEEE International Conference on Robotics and Automation (ICRA), page 6672–6679. IEEE, 2024

    2024

  12. [12]

    Droid: A large-scale in-the-wild robot manipulation dataset

    Alexander Khazatsky, Karl Pertsch, Suraj Nair, Ashwin Balakrishna, Sudeep Dasari, Siddharth Karam- cheti, Soroush Nasiriany, Mohan Srirama, Lawrence Chen, Kirsty Ellis, Peter Fagan, Joey Hejna, Masha Itkina, Marion Lepert, Yecheng Ma, Patrick Miller, Jimmy Wu, Suneel Belkhale, Shivin Dass, Huy Ha, Arhan Jain, Abraham Lee, Youngwoon Lee, Marius Memmel, Sun...

    2024

  13. [13]

    Bfm-zero: A promptable behavioral foundation model for humanoid control using unsupervised reinforcement learning, 2025

    Yitang Li, Zhengyi Luo, Tonghe Zhang, Cunxi Dai, Anssi Kanervisto, Andrea Tirinzoni, Haoyang Weng, Kris Kitani, Mateusz Guzek, Ahmed Touati, Alessandro Lazaric, Matteo Pirotta, and Guanya Shi. Bfm-zero: A promptable behavioral foundation model for humanoid control using unsupervised reinforcement learning, 2025

    2025

  14. [14]

    Lychsim: A controllable and interactive simulation framework for vision research.arXiv preprint arXiv:2605.12449, 2026

    Wufei Ma, Chloe Wang, Siyi Chen, Jiawei Peng, Patrick Li, and Alan Yuille. Lychsim: A controllable and interactive simulation framework for vision research.arXiv preprint arXiv:2605.12449, 2026

    Pith/arXiv arXiv 2026

  15. [15]

    Scalable real2sim: Physics- aware asset generation via robotic pick-and-place setups

    Nicholas Pfaff, Evelyn Fu, Jeremy Binagia, Phillip Isola, and Russ Tedrake. Scalable real2sim: Physics- aware asset generation via robotic pick-and-place setups. In2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), page 6296–6303. IEEE, 2025

    2025

  16. [16]

    Scenesmith: Agentic generation of simulation-ready indoor scenes.arXiv preprint arXiv:2602.09153, 2026

    Nicholas Pfaff, Thomas Cohn, Sergey Zakharov, Rick Cory, and Russ Tedrake. Scenesmith: Agentic generation of simulation-ready indoor scenes.arXiv preprint arXiv:2602.09153, 2026. 11 Agentic Real2Sim: Generalized Physical World Modeling with Vision-Language Agents

    Pith/arXiv arXiv 2026

  17. [17]

    Simfoundry: Modular and automated scene generation for policy learning and evaluation.arXiv preprint arXiv:2606.28276, 2026

    Nadun Ranawaka, Josiah Wong, Wei-Lin Pai, Wei-Teng Chu, Tianyuan Dai, Masoud Moghani, Hang Yin, Yunfan Jiang, Wesley Durbano, Brandon Huynh, et al. Simfoundry: Modular and automated scene generation for policy learning and evaluation.arXiv preprint arXiv:2606.28276, 2026

    Pith/arXiv arXiv 2026

  18. [18]

    Mujoco: A physics engine for model-based control

    Emanuel Todorov, Tom Erez, and Yuval Tassa. Mujoco: A physics engine for model-based control. In 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, page 5026–5033. IEEE, 2012

    2012

  19. [19]

    Lodestar: Long-horizon dexterity via synthetic data augmentation from human demonstrations

    Weikang Wan, Jiawei Fu, Xiaodi Yuan, Yifeng Zhu, and Hao Su. Lodestar: Long-horizon dexterity via synthetic data augmentation from human demonstrations. In9th Annual Conference on Robot Learning, 2025

    2025

  20. [20]

    Physcensis: Physics-augmented llm agents for complex physical scene arrangement.arXiv preprint arXiv:2602.14968, 2026

    Yian Wang, Han Yang, Minghao Guo, Xiaowen Qiu, Tsun-Hsuan Wang, Wojciech Matusik, Joshua B Tenenbaum, and Chuang Gan. Physcensis: Physics-augmented llm agents for complex physical scene arrangement.arXiv preprint arXiv:2602.14968, 2026

    arXiv 2026

  21. [21]

    Foundationpose: Unified 6d pose estimation and tracking of novel objects

    Bowen Wen, Wei Yang, Jan Kautz, and Stan Birchfield. Foundationpose: Unified 6d pose estimation and tracking of novel objects. In2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), page 17868–17879. IEEE, 2024

    2024

  22. [22]

    Founda- tionstereo: Zero-shot stereo matching.CVPR, 2025

    Bowen Wen, Matthew Trepte, Joseph Aribido, Jan Kautz, Orazio Gallo, and Stan Birchfield. Founda- tionstereo: Zero-shot stereo matching.CVPR, 2025

    2025

  23. [23]

    Holodeck: Language guided generation of 3d embodied ai environments

    Yue Yang, Fan-Yun Sun, Luca Weihs, Eli Vanderbilt, Alvaro Herrasti, Winson Han, Jiajun Wu, Nick Haber, Ranjay Krishna, Lingjie Liu, Chris Callison-Burch, Mark Yatskar, Aniruddha Kembhavi, and Christopher Clark. Holodeck: Language guided generation of 3d embodied ai environments. In2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),...

    2024

  24. [24]

    Sceneweaver: All-in-one 3d scene synthesis with an extensible and self-reflective agent.Advances in neural information processing systems, 38: 140319–140351, 2026

    Yandan Yang, Baoxiong Jia, Shujie Zhang, and Siyuan Huang. Sceneweaver: All-in-one 3d scene synthesis with an extensible and self-reflective agent.Advances in neural information processing systems, 38: 140319–140351, 2026

    2026

  25. [25]

    Articraft: An agentic system for scalable articulated 3d asset generation.arXiv preprint arXiv:2605.15187, 2026

    Matt Zhou, Ruining Li, Xiaoyang Lyu, Zhaomou Song, Zhening Huang, Chuanxia Zheng, Christian Rupprecht, Andrea Vedaldi, and Shangzhe Wu. Articraft: An agentic system for scalable articulated 3d asset generation.arXiv preprint arXiv:2605.15187, 2026

    Pith/arXiv arXiv 2026

  26. [26]

    Grs: Generating robotic simulation tasks from real-world images

    Alex Zook, Fan-Yun Sun, Josef Spjut, Valts Blukis, Stan Birchfield, and Jonathan Tremblay. Grs: Generating robotic simulation tasks from real-world images. In2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPR W), page 594–603. IEEE, 2025. 12

    2025