Pith. sign in

REVIEW 2 major objections 1 minor 5 cited by

Articraft reduces articulated 3D asset creation to LLM-written programs guided by a domain SDK and validation harness.

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 · grok-4.3

2026-06-30 20:47 UTC pith:NICLPOVG

load-bearing objection Articraft reduces articulated asset creation to LLM code-writing against a custom SDK and validation harness, then claims a 10K dataset, but the abstract shows no numbers on success rates or quality metrics. the 2 major comments →

arxiv 2605.15187 v1 pith:NICLPOVG submitted 2026-05-14 cs.CV cs.GRcs.RO

Articraft: An Agentic System for Scalable Articulated 3D Asset Generation

Matt Zhou , Ruining Li , Xiaoyang Lyu , Zhaomou Song , Zhening Huang , Chuanxia Zheng , Christian Rupprecht , Andrea Vedaldi
show 1 more author
Shangzhe Wu
This is my paper
classification cs.CV cs.GRcs.RO
keywords articulated 3D assetsLLM agents3D asset generationsynthetic datasetsURDFrobotics simulationvirtual reality
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 targets the shortage of large, diverse datasets needed to train models that understand articulated 3D objects. It reframes generation as the task of writing a program that defines parts, assembles geometry, specifies joints, and includes validation tests. Articraft supplies the LLM with a restricted workspace, a domain-specific SDK, and structured feedback from automated validation so the model avoids low-level file and environment management. When evaluated, this yields higher-quality assets than both specialized articulated generators and general coding agents. The system also produces a curated dataset of more than 10,000 assets across 245 categories that supports downstream training and applications.

Core claim

Articraft reduces the generation of articulated 3D assets to the task of writing a program that builds the asset. An LLM writes code against a domain-specific SDK that handles part definition, geometry composition, joint specification, and test writing. A harness restricts the workspace, validates the resulting assets, and returns structured feedback, freeing the LLM from authoring URDF files or managing software environments. This produces higher-quality assets than state-of-the-art articulated-asset generators and general-purpose coding agents, and enables the construction of a dataset of over 10,000 assets spanning 245 categories.

What carries the argument

The domain-specific SDK and validation harness that together guide the LLM to write, test, and iterate on articulated-asset programs inside a restricted workspace.

Load-bearing premise

An LLM guided only by the provided programmatic SDK, restricted workspace, and structured validation feedback can reliably produce valid, high-quality articulated asset programs without substantial human intervention or post-generation fixes.

What would settle it

Run Articraft on a fixed set of prompts, then measure the fraction of outputs that fail basic joint validity checks or receive lower human preference scores than assets from baseline generators.

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

If this is right

  • Produces higher-quality articulated assets than existing generators and coding agents.
  • Enables construction of a curated dataset containing over 10,000 assets in 245 categories.
  • Supports training of models that learn articulated asset structure.
  • Facilitates downstream uses in robotics simulation and virtual reality.

Where Pith is reading between the lines

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

  • The same harness structure could be reused with stronger future LLMs to increase asset diversity without redesigning the SDK.
  • The programmatic reduction may generalize to other structured 3D generation problems that require precise part relationships.
  • Automated validation loops could lower the cost of curating even larger synthetic datasets for simulation training.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 1 minor

Summary. The paper claims that Articraft, an agentic system using LLMs to write programs for articulated 3D assets via a domain-specific SDK and validation harness, generates higher-quality assets than state-of-the-art articulated-asset generators and general-purpose coding agents. It further claims to have used this system to create the Articraft-10K dataset of over 10K articulated assets across 245 categories, which is useful for training models and in applications like robotics simulation and virtual reality.

Significance. If validated, this approach could significantly alleviate the data scarcity issue in articulated 3D object understanding by enabling scalable, automated generation of diverse assets. The agentic harness design, which provides structured feedback without requiring the LLM to handle low-level details like URDF files, represents a promising direction for LLM-based asset generation if the reliability claims are substantiated with quantitative evidence.

major comments (2)
  1. [Abstract] Abstract: The claim that Articraft produces higher-quality assets than both state-of-the-art articulated-asset generators and general-purpose coding agents is asserted without any supporting metrics, baselines, quantitative results, or description of the evaluation protocol.
  2. [Abstract] Abstract: No data is provided on per-asset success rates, average number of LLM iterations, failure modes, or the extent of human intervention required, which is critical to substantiate the scalability to 10K assets across 245 categories and the reliability of the LLM guided by the SDK and harness.
minor comments (1)
  1. The abstract could benefit from a brief mention of the specific evaluation metrics used for quality assessment.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the abstract. The comments correctly identify that the abstract makes claims without embedding the supporting quantitative details. We will revise the abstract to include key metrics, a brief description of the evaluation protocol, and statistics on system reliability to make the claims self-contained. We address each point below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that Articraft produces higher-quality assets than both state-of-the-art articulated-asset generators and general-purpose coding agents is asserted without any supporting metrics, baselines, quantitative results, or description of the evaluation protocol.

    Authors: The full manuscript provides quantitative comparisons against baselines (both articulated-asset generators and general coding agents) along with the evaluation protocol in Sections 4 and 5. The abstract summarizes these findings at a high level. We agree that the abstract would be stronger if it included a small number of key metrics and a one-sentence reference to the evaluation protocol; we will revise the abstract accordingly. revision: yes

  2. Referee: [Abstract] Abstract: No data is provided on per-asset success rates, average number of LLM iterations, failure modes, or the extent of human intervention required, which is critical to substantiate the scalability to 10K assets across 245 categories and the reliability of the LLM guided by the SDK and harness.

    Authors: The manuscript describes the SDK, harness, and iterative generation process in Section 3 and reports overall success in producing the 10K-asset dataset. However, the abstract does not include the requested per-asset statistics. We will add a concise summary of success rates, average iterations, main failure modes, and human intervention levels (drawn from our logged experiments) directly into the revised abstract. revision: yes

Circularity Check

0 steps flagged

No circularity: claims rest on external comparisons and system description

full rationale

The paper presents an engineering system (Articraft) that uses LLMs with a domain-specific SDK, restricted workspace, and validation harness to generate articulated 3D asset programs. The central claims—that the system yields higher-quality assets than SOTA generators and coding agents, and enables a 10K-asset dataset—are supported by external comparisons and downstream utility demonstrations rather than any internal mathematical derivation, fitted parameter renamed as prediction, or self-citation chain. No equations, uniqueness theorems, or ansatzes appear in the provided text that reduce to inputs by construction. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Central claim rests on the premise that LLMs can be steered by the described interface and feedback loop to produce valid articulated assets at scale; this is an unverified domain assumption given only the abstract.

axioms (1)
  • domain assumption LLMs can be effectively guided by a domain-specific SDK, restricted workspace, and structured validation feedback to generate valid articulated 3D asset programs.
    This assumption is required for the agentic system to succeed and is not independently supported in the provided abstract.

pith-pipeline@v0.9.1-grok · 5790 in / 1205 out tokens · 36582 ms · 2026-06-30T20:47:59.033029+00:00 · methodology

0 comments
read the original abstract

A bottleneck in learning to understand articulated 3D objects is the lack of large and diverse datasets. In this paper, we propose to leverage large language models (LLMs) to close this gap and generate articulated assets at scale. We reduce the problem of generating an articulated 3D asset to that of writing a program that builds it. We then introduce a new agentic system, Articraft, that writes such programs automatically. We design a programmatic interface and harness to help the LLM do so effectively. The LLM writes code against a domain-specific SDK for defining parts, composing geometry, specifying joints, and writing tests to validate the resulting assets. The harness exposes a restricted workspace and interface to the LLM, validates the resulting assets, and returns structured feedback. In this way, the LLM is not distracted by details such as authoring a URDF file or managing a complex software environment. We show that this produces higher-quality assets than both state-of-the-art articulated-asset generators and general-purpose coding agents. Using Articraft, we build Articraft-10K, a curated dataset of over 10K articulated assets spanning 245 categories, and show its utility both for training models of articulated assets and in downstream applications such as robotics simulation and virtual reality.

Figures

Figures reproduced from arXiv: 2605.15187 by Andrea Vedaldi, Christian Rupprecht, Chuanxia Zheng, Matt Zhou, Ruining Li, Shangzhe Wu, Xiaoyang Lyu, Zhaomou Song, Zhening Huang.

Figure 1
Figure 1. Figure 1: We introduce Articraft, an agentic system for scalable articulated 3D asset generation. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Excerpt from a model.py program authored by Articraft for an articulated desk lamp. 3.1 The Articraft Programmatic Interface Each 3D asset a generated by Articraft is defined by a program y consisting of a single Python file, model.py (see the example in [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Examples of articulated 3D assets in the Articraft-10K dataset generated by Articraft. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of articulated asset generation with different methods. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Comparison of 3D articulated asset generation with different base models, demonstrating [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 6
Figure 6. Figure 6: User study results. Results. The user study results are summa￾rized in [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Image-conditioned generation results from Articraft. Given a reference image, Articraft [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The articulated 3D assets generated by Articraft can be directly imported into the physics [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Example of the Articraft SDK’s expressivity and style controllability. The same block [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Statistics of Articraft-10K: (a) Number of objects in each super-category. (b) Average cost [PITH_FULL_IMAGE:figures/full_fig_p015_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Word cloud of object names in Articraft-10K. [PITH_FULL_IMAGE:figures/full_fig_p015_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Distribution of cost, number of turns, and number of links per object in Articraft-10K. [PITH_FULL_IMAGE:figures/full_fig_p016_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Mesh statistics of Articraft-10K. outputs are merged back into LiteReality’s parsed scene-layout integration stage, yielding a room that closely matches the captured imagery while being fully articulated and ready for simulation and downstream interaction tasks, as shown in [PITH_FULL_IMAGE:figures/full_fig_p016_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Distribution of URDF joints per object in Articraft-10K. [PITH_FULL_IMAGE:figures/full_fig_p017_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: More examples of image-conditioned articulated 3D asset generations produced by [PITH_FULL_IMAGE:figures/full_fig_p017_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Articraft integrates seamlessly into the LiteReality [ [PITH_FULL_IMAGE:figures/full_fig_p018_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Examples of failure cases of Articraft. Finally, some failures are sporadic in more complex categories. The agent may omit interior structure or fail to hollow out shapes even when the exterior and articulation are plausible, as in “rice cooker” and “refrigerator with hinged doors” cases These errors reflect the current tradeoff between cheap validation and stronger semantic or functional checks: the harn… view at source ↗
Figure 19
Figure 19. Figure 19: Additional articulated assets generated by Articraft. The results span a broad range [PITH_FULL_IMAGE:figures/full_fig_p020_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Curated excerpt from a GPT-5.5 trace for a 5-star rolling toolbox asset. The trace shows [PITH_FULL_IMAGE:figures/full_fig_p025_20.png] view at source ↗

discussion (0)

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

Forward citations

Cited by 5 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. UnfoldArt: Zero-Shot Recovery of Full Articulated 3D Objects from Text or Image

    cs.CV 2026-06 unverdicted novelty 7.0

    UnfoldArt uses multi-agent debate grounded in vision-language and video models to infer articulation parameters and reconstruct full 3D objects including occluded parts from text or image inputs.

  2. UnfoldArt: Zero-Shot Recovery of Full Articulated 3D Objects from Text or Image

    cs.CV 2026-06 unverdicted novelty 7.0

    UnfoldArt uses a two-round structured debate between high-level semantic agents and low-level parameter agents, grounded in generated video, to infer articulation and reconstruct full articulated 3D objects including ...

  3. P3D-Bench: Benchmarking MLLMs for Parametric 3D Generation and Structural Reasoning

    cs.CV 2026-06 unverdicted novelty 7.0

    P3D-Bench is a benchmark with three task families that scores MLLMs on generating executable parametric 3D programs, finding failures in precise geometry and part assembly.

  4. 3DCodeBench: Benchmarking Agentic Procedural 3D Modeling Via Code

    cs.CV 2026-05 unverdicted novelty 7.0

    3DCodeBench is a new benchmark evaluating 12 VLMs on translating multimodal prompts into procedural 3D modeling code, paired with 3DCodeArena for human preference rankings.

  5. Sequential Planning via Anchored Robotic Keypoints

    cs.RO 2026-06 unverdicted novelty 6.0

    SPARK reaches 43.7% success on six LIBERO-PRO cells by LLM-generated typed behavior trees plus multi-prompt perception and recovery, more than doubling CaP-Agent0 and VLA baselines.

Reference graph

Works this paper leans on

65 extracted references · 3 canonical work pages · cited by 4 Pith papers · 1 internal anchor

  1. [1]

    Claude Code

    Anthropic. Claude Code. https://docs.anthropic.com/en/docs/claude-code/overview, 2026. [Online; Accessed 06-May-2026]

    2026

  2. [2]

    Introducing Claude Opus 4.7.https://www.anthropic.com/news/claude-opus-4-7,

    Anthropic. Introducing Claude Opus 4.7.https://www.anthropic.com/news/claude-opus-4-7,

  3. [3]

    [Online; Accessed 06-May-2026]

    2026

  4. [4]

    PhysX-3D: physical-grounded 3D asset generation

    Ziang Cao, Zhaoxi Chen, Liang Pan, and Ziwei Liu. PhysX-3D: physical-grounded 3D asset generation. InNeurIPS, 2025

    2025

  5. [5]

    PhysX-Anything: simulation-ready physical 3D assets from single image

    Ziang Cao, Fangzhou Hong, Zhaoxi Chen, Liang Pan, and Ziwei Liu. PhysX-Anything: simulation-ready physical 3D assets from single image. InCVPR, 2026

    2026

  6. [6]

    AutoPartGen: Autoregressive 3D part generation and discovery

    Minghao Chen, Jianyuan Wang, Roman Shapovalov, Tom Monnier, Hyunyoung Jung, Dilin Wang, Rakesh Ranjan, Iro Laina, and Andrea Vedaldi. AutoPartGen: Autoregressive 3D part generation and discovery. In NeurIPS, 2025

    2025

  7. [7]

    URDFormer: A pipeline for constructing articulated simulation environments from real-world images

    Zoey Chen, Aaron Walsman, Marius Memmel, Kaichun Mo, Alex Fang, Karthikeya Vemuri, Alan Wu, Dieter Fox, and Abhishek Gupta. URDFormer: A pipeline for constructing articulated simulation environments from real-world images. InRSS, 2024

    2024

  8. [8]

    Cadquery, 2026

    CadQuery contributors. Cadquery, 2026

    2026

  9. [9]

    Make-it-Real: Unleashing large multimodal model for painting 3D objects with realistic materials

    Ye Fang, Zeyi Sun, Tong Wu, Jiaqi Wang, Ziwei Liu, Gordon Wetzstein, and Dahua Lin. Make-it-Real: Unleashing large multimodal model for painting 3D objects with realistic materials. InNeurIPS, 2024

    2024

  10. [10]

    MeshArt: Generating articulated meshes with structure-guided transformers

    Daoyi Gao, Yawar Siddiqui, Lei Li, and Angela Dai. MeshArt: Generating articulated meshes with structure-guided transformers. InCVPR, 2025

    2025

  11. [11]

    PAL: Program-aided language models

    Luyu Gao, Aman Madaan, Shuyan Zhou, Uri Alon, Pengfei Liu, Yiming Yang, Jamie Callan, and Graham Neubig. PAL: Program-aided language models. InICML, 2023

    2023

  12. [12]

    GAPartNet: Cross-category domain-generalizable object perception and manipulation via generalizable and actionable parts

    Haoran Geng, Helin Xu, Chengyang Zhao, Chao Xu, Li Yi, Siyuan Huang, and He Wang. GAPartNet: Cross-category domain-generalizable object perception and manipulation via generalizable and actionable parts. InCVPR, 2023

    2023

  13. [13]

    Gemini 3.1 Pro Model Card

    Google DeepMind. Gemini 3.1 Pro Model Card. https://deepmind.google/models/model-cards/gemini-3-1-pro/, 2026. [Online; Accessed 06-May-2026]

    2026

  14. [14]

    CAD-coder: Text-to-CAD generation with chain-of-thought and geometric reward

    Yandong Guan, Xilin Wang, XiMing Xing, Jing Zhang, Dong Xu, and Qian Yu. CAD-coder: Text-to-CAD generation with chain-of-thought and geometric reward. InNeurIPS, 2025

    2025

  15. [15]

    LiteReality: Graphics-ready 3D scene reconstruction from RGB-D scans

    Zhening Huang, Xiaoyang Wu, Fangcheng Zhong, Hengshuang Zhao, Matthias Nießner, and Joan Lasenby. LiteReality: Graphics-ready 3D scene reconstruction from RGB-D scans. InNeurIPS, 2025

    2025

  16. [16]

    Ditto: Building digital twins of articulated objects from interaction

    Zhenyu Jiang, Cheng-Chun Hsu, and Yuke Zhu. Ditto: Building digital twins of articulated objects from interaction. InCVPR, 2022

    2022

  17. [17]

    Kenny Jones, Theresa Barton, Xianghao Xu, Kai Wang, Ellen Jiang, Paul Guerrero, Niloy J

    R. Kenny Jones, Theresa Barton, Xianghao Xu, Kai Wang, Ellen Jiang, Paul Guerrero, Niloy J. Mitra, and Daniel Ritchie. ShapeAssembly: Learning to generate programs for 3D shape structure synthesis. In SIGGRAPH, 2020

    2020

  18. [18]

    Kenny Jones, David Charatan, Paul Guerrero, Niloy J

    R. Kenny Jones, David Charatan, Paul Guerrero, Niloy J. Mitra, and Daniel Ritchie. ShapeMOD: Macro operation discovery for 3D shape programs. InSIGGRAPH, 2021. 10

    2021

  19. [19]

    Infinigen-Sim: Procedural generation of articulated simulation assets.arXiv, 2505.10755, 2025

    Abhishek Joshi, Beining Han, Jack Nugent, Yiming Zuo, Jonathan Liu, Hongyu Wen, Stamatis Alexandropoulos, Tao Sun, Alexander Raistrick, Gaowen Liu, Yi Shao, and Jia Deng. Infinigen-Sim: Procedural generation of articulated simulation assets.arXiv, 2505.10755, 2025

    work page arXiv 2025

  20. [20]

    Text2CAD: Generating sequential CAD designs from beginner-to-expert level text prompts

    Mohammad Sadil Khan, Sankalp Sinha, Sheikh Talha Uddin, Didier Stricker, Sk Aziz Ali, and Muhammad Zeshan Afzal. Text2CAD: Generating sequential CAD designs from beginner-to-expert level text prompts. InNeurIPS, 2024

    2024

  21. [21]

    Articulate-Anything: Automatic modeling of articulated objects via a vision-language foundation model

    Long Le, Jason Xie, William Liang, Hung-Ju Wang, Yue Yang, Yecheng Jason Ma, Kyle Vedder, Arjun Krishna, Dinesh Jayaraman, and Eric Eaton. Articulate-Anything: Automatic modeling of articulated objects via a vision-language foundation model. InICLR, 2025

    2025

  22. [22]

    Guibas, and Kostas Daniilidis

    Jiahui Lei, Congyue Deng, Bokui Shen, Leonidas J. Guibas, and Kostas Daniilidis. NAP: Neural 3D articulation prior. InNeurIPS, 2023

    2023

  23. [23]

    Particulate: Feed-forward 3D object articulation

    Ruining Li, Yuxin Yao, Chuanxia Zheng, Christian Rupprecht, Joan Lasenby, Shangzhe Wu, and Andrea Vedaldi. Particulate: Feed-forward 3D object articulation. InCVPR, 2026

    2026

  24. [24]

    Guibas, A

    Xiaolong Li, He Wang, Li Yi, Leonidas J. Guibas, A. Lynn Abbott, and Shuran Song. Category-level articulated object pose estimation. InCVPR, 2020

    2020

  25. [25]

    URDF-Anything: Constructing articulated objects with 3D multimodal language model

    Zhe Li, Xiang Bai, Jieyu Zhang, Zhuangzhe Wu, Che Xu, Ying Li, Chengkai Hou, and Shanghang Zhang. URDF-Anything: Constructing articulated objects with 3D multimodal language model. InNeurIPS, 2025

    2025

  26. [26]

    Code as policies: Language model programs for embodied control

    Jacky Liang, Wenlong Huang, Fei Xia, Peng Xu, Karol Hausman, Brian Ichter, Pete Florence, and Andy Zeng. Code as policies: Language model programs for embodied control. InICRA, 2023

    2023

  27. [27]

    Simready: Simulation-ready 3d assets.https://simready.com/, 2025

    Lightwheel. Simready: Simulation-ready 3d assets.https://simready.com/, 2025. Accessed: 2025

    2025

  28. [28]

    PARIS: Part-level reconstruction and motion analysis for articulated objects

    Jiayi Liu, Ali Mahdavi-Amiri, and Manolis Savva. PARIS: Part-level reconstruction and motion analysis for articulated objects. InICCV, 2023

    2023

  29. [29]

    CAGE: Controllable Articulation GEneration

    Jiayi Liu, Hou In Ivan Tam, Ali Mahdavi-Amiri, and Manolis Savva. CAGE: Controllable Articulation GEneration. InCVPR, 2024

    2024

  30. [30]

    Chang, Manolis Savva, and Ali Mahdavi-Amiri

    Jiayi Liu, Denys Iliash, Angel X. Chang, Manolis Savva, and Ali Mahdavi-Amiri. SINGAPO: Single image controlled generation of articulated parts in objects. InICLR, 2025

    2025

  31. [31]

    Survey on modeling of human-made articulated objects

    Jiayi Liu, Manolis Savva, and Ali Mahdavi-Amiri. Survey on modeling of human-made articulated objects. Computer Graphics F orum, 2025

    2025

  32. [32]

    AKB-48: A real-world articulated object knowledge base

    Liu Liu, Wenqiang Xu, Haoyuan Fu, Sucheng Qian, Qiaojun Yu, Yang Han, and Cewu Lu. AKB-48: A real-world articulated object knowledge base. InCVPR, 2022

    2022

  33. [33]

    Chang, Li Yi, Subarna Tripathi, Leonidas J

    Kaichun Mo, Shilin Zhu, Angel X. Chang, Li Yi, Subarna Tripathi, Leonidas J. Guibas, and Hao Su. PartNet: A large-scale benchmark for fine-grained and hierarchical part-level 3D object understanding. In CVPR, 2019

    2019

  34. [34]

    A-SDF: Learning disentangled signed distance functions for articulated shape representation

    Jiteng Mu, Weichao Qiu, Adam Kortylewski, Alan Yuille, Nuno Vasconcelos, and Xiaolong Wang. A-SDF: Learning disentangled signed distance functions for articulated shape representation. InICCV, 2021

    2021

  35. [35]

    RoboCasa365: A large-scale simulation framework for training and benchmarking generalist robots

    Soroush Nasiriany, Sepehr Nasiriany, Abhiram Maddukuri, and Yuke Zhu. RoboCasa365: A large-scale simulation framework for training and benchmarking generalist robots. InICLR, 2026

    2026

  36. [36]

    Isaac sim, 2025

    NVIDIA. Isaac sim, 2025

    2025

  37. [37]

    Introducing GPT-5.2.https://openai.com/index/introducing-gpt-5-2/, 2025

    OpenAI. Introducing GPT-5.2.https://openai.com/index/introducing-gpt-5-2/, 2025. [Online; Accessed 06-May-2026]

    2025

  38. [38]

    Introducing GPT-5.2-Codex

    OpenAI. Introducing GPT-5.2-Codex. https://openai.com/index/introducing-gpt-5-2-codex/, 2025. [Online; Accessed 06-May-2026]

    2025

  39. [39]

    Codex CLI.https://developers.openai.com/codex/cli, 2026

    OpenAI. Codex CLI.https://developers.openai.com/codex/cli, 2026. [Online; Accessed 06-May-2026]

    2026

  40. [40]

    Introducing GPT-5.3-Codex

    OpenAI. Introducing GPT-5.3-Codex. https://openai.com/index/introducing-gpt-5-3-codex/, 2026. [Online; Accessed 06-May-2026]. 11

    2026

  41. [41]

    Introducing GPT-5.4.https://openai.com/index/introducing-gpt-5-4/, 2026

    OpenAI. Introducing GPT-5.4.https://openai.com/index/introducing-gpt-5-4/, 2026. [Online; Accessed 06-May-2026]

    2026

  42. [42]

    Introducing GPT-5.5.https://openai.com/index/introducing-gpt-5-5/, 2026

    OpenAI. Introducing GPT-5.5.https://openai.com/index/introducing-gpt-5-5/, 2026. [Online; Accessed 06-May-2026]

    2026

  43. [43]

    Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy V . V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel HAZIZA, Francisco Massa, Alaaeldin El-Nouby, Mido Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Herve Jegou, Julien Mairal, Patrick L...

    2024

  44. [44]

    Infinite photorealistic worlds using procedural generation

    Alexander Raistrick, Lahav Lipson, Zeyu Ma, Lingjie Mei, Mingzhe Wang, Yiming Zuo, Karhan Kayan, Hongyu Wen, Beining Han, Yihan Wang, Alejandro Newell, Hei Law, Ankit Goyal, Kaiyu Yang, and Jia Deng. Infinite photorealistic worlds using procedural generation. InCVPR, 2023

    2023

  45. [45]

    Reflexion: Language agents with verbal reinforcement learning

    Noah Shinn, Federico Cassano, Edward Berman, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. Reflexion: Language agents with verbal reinforcement learning. InNeurIPS, 2023

    2023

  46. [46]

    ArtiCAD: Articulated CAD Assembly Design via Multi-Agent Code Generation

    Yuan Shui, Yandong Guan, Zhanwei Zhang, Juncheng Hu, Jing Zhang, Dong Xu, and Qian Yu. Articad: Articulated cad assembly design via multi-agent code generation.arXiv preprint arXiv:2604.10992, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  47. [47]

    Denoising diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. InICLR, 2021

    2021

  48. [48]

    ArtFormer: controllable generation of diverse 3d articulated objects

    Jiayi Su, Youhe Feng, Zheng Li, Jinhua Song, Yangfan He, Botao Ren, and Botian Xu. ArtFormer: controllable generation of diverse 3d articulated objects. InCVPR, 2025

    2025

  49. [49]

    Matsynth: A modern pbr materials dataset

    Giuseppe Vecchio and Valentin Deschaintre. Matsynth: A modern pbr materials dataset. InCVPR, 2024

    2024

  50. [50]

    V oyager: An open-ended embodied agent with large language models.Transactions on Machine Learning Research, 2025

    Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. V oyager: An open-ended embodied agent with large language models.Transactions on Machine Learning Research, 2025

    2025

  51. [51]

    GRUtopia: Dream general robots in a city at scale.arXiv, 2024

    Hanqing Wang, Jiahe Chen, Wensi Huang, Qingwei Ben, Tai Wang, Boyu Mi, Tao Huang, Siheng Zhao, Yilun Chen, Sizhe Yang, Peizhou Cao, Wenye Yu, Zichao Ye, Jialun Li, Junfeng Long, Zirui Wang, Huiling Wang, Ying Zhao, Zhongying Tu, Yu Qiao, Dahua Lin, and Jiangmiao Pang. GRUtopia: Dream general robots in a city at scale.arXiv, 2024

    2024

  52. [52]

    Shape2Motion: Joint analysis of motion parts and attributes from 3D shapes

    Xiaogang Wang, Bin Zhou, Yahao Shi, Xiaowu Chen, Qinping Zhao, and Kai Xu. Shape2Motion: Joint analysis of motion parts and attributes from 3D shapes. InCVPR, 2019

    2019

  53. [53]

    Yijia Weng, He Wang, Qiang Zhou, Yuzhe Qin, Yueqi Duan, Qingnan Fan, Baoquan Chen, Hao Su, and Leonidas J. Guibas. CAPTRA: Category-level pose tracking for rigid and articulated objects from point clouds. InICCV, 2021

    2021

  54. [54]

    DeepCAD: a deep generative network for computer-aided design models

    Rundi Wu, Chang Xiao, and Changxi Zheng. DeepCAD: a deep generative network for computer-aided design models. InICCV, 2021

    2021

  55. [55]

    URDF-Anything+: Autoregressive articulated 3D models generation for physical simulation.arXiv, 2026

    Zhuangzhe Wu, Yue Xin, Chengkai Hou, Minghao Chen, Yaoxu Lyu, Jieyu Zhang, and Shanghang Zhang. URDF-Anything+: Autoregressive articulated 3D models generation for physical simulation.arXiv, 2026

    2026

  56. [56]

    Chang, Leonidas J

    Fanbo Xiang, Yuzhe Qin, Kaichun Mo, Yikuan Xia, Hao Zhu, Fangchen Liu, Minghua Liu, Hanxiao Jiang, Yifu Yuan, He Wang, Li Yi, Angel X. Chang, Leonidas J. Guibas, and Hao Su. SAPIEN: A simulated part-based interactive environment. InCVPR, 2020

    2020

  57. [57]

    Structured 3D latents for scalable and versatile 3d generation

    Jianfeng Xiang, Zelong Lv, Sicheng Xu, Yu Deng, Ruicheng Wang, Bowen Zhang, Dong Chen, Xin Tong, and Jiaolong Yang. Structured 3D latents for scalable and versatile 3d generation. InCVPR, 2025

    2025

  58. [58]

    Text-to-CadQuery: A new paradigm for CAD generation with scalable large model capabilities.arXiv, 2025

    Haoyang Xie and Feng Ju. Text-to-CadQuery: A new paradigm for CAD generation with scalable large model capabilities.arXiv, 2025

    2025

  59. [59]

    Jimenez, Alexander Wettig, Kilian Lieret, Shunyu Yao, Karthik R

    John Yang, Carlos E. Jimenez, Alexander Wettig, Kilian Lieret, Shunyu Yao, Karthik R. Narasimhan, and Ofir Press. Swe-agent: Agent-computer interfaces enable automated software engineering. InNeurIPS, 2024

    2024

  60. [60]

    ReAct: Synergizing reasoning and acting in language models

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. ReAct: Synergizing reasoning and acting in language models. InICLR, 2023

    2023

  61. [61]

    game boy

    Zhao Mandi, Yijia Weng, Dominik Bauer, and Shuran Song. Real2Code: Reconstruct Articulated Objects via Code Generation. InICLR, 2025. 12 A Additional Statistics A.1 Articraft-10K Dataset Figure 11a shows the distribution of objects across the 15 super-categories in Articraft-10K. Fig- ure 11b plots the average cost of each category against the average num...

    work page arXiv 2025

  62. [62]

    Choose the SDK r e p r e s e n t a t i o n that best matches the real form

    R E A L I S T I C G EOM ET RY --- this is the do mi na nt quality bar . Choose the SDK r e p r e s e n t a t i o n that best matches the real form . Use simple p r i m i t i v e s when they are g e n u i n e l y correct ; use lofts , sweeps , booleans , wires , or C ad Qu er y ge om etr y when the shape needs them . Objects that are hollow in reality ( cu...

  63. [63]

    Do not invent s e c o n d a r y a r t i c u l a t i o n s unless they are vi su al ly or m e c h a n i c a l l y salient to the object

    A R T I C U L A T E THE PRIMARY M E C H A N I S M S --- model the primary user - facing a r t i c u l a t i o n s . Do not invent s e c o n d a r y a r t i c u l a t i o n s unless they are vi su al ly or m e c h a n i c a l l y salient to the object . On appliances , electronics , instruments , and other control - heavy objects , buttons , knobs , switch...

  64. [64]

    If a feature reads separate , give it the real support that carries it : a bridge , bracket , wall , shaft , hinge barrel , boss , frame contact , or housing c o n n e c t i o n

    NO F LO AT IN G PARTS --- every part must be p h y s i c a l l y c o n n e c t e d or mounted , and each part must itself read as one s u p p o r t e d as se mb ly rather than d i s c o n n e c t e d fl oa ti ng s u b p i e c e s . If a feature reads separate , give it the real support that carries it : a bridge , bracket , wall , shaft , hinge barrel , b...

  65. [65]

    model . py

    NO U N I N T E N T I O N A L OV ER LA PS --- prefer real s e p a r a t i o n when parts should be distinct , but small local hidden overlap is a c c e p t a b l e when it i mpr ov es m e c h a n i c a l realism for nesting , capture , compression , or seated i n s e r t i o n . Keep i n t e n t i o n a l overlap local and element - scoped when possible , ...

    2026