Pith. sign in

REVIEW 3 major objections 4 minor 3 cited by

ScrewSplat: An End-to-End Method for Articulated Object Recognition

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read ScrewSplat claims that a single RGB video is enough to reconstruct an object's 3D geometry, segment it into rigid movable parts, and recover each part's screw axis, enabling zero-shot text-guided manipulation.

desk verdict A genuinely new RGB-only articulation-recovery idea that deserves referee time, but this corrupted copy makes the soundness claims unverifiable. read the letter →

arxiv 2508.02146 v2 pith:5T6QDCEP submitted 2025-08-04 cs.RO cs.CV

classification cs.ROcs.CV
keywords articulatedobjectrecognitionscrewaxesGaussianSplattingend-to-endlearningRGB-onlyperceptionkinematicstructurerecoveryzero-shotmanipulationrobot
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

ScrewSplat tries to establish that articulated object recognition does not need depth cameras, a known number of parts, or a chain of separate processing stages. The paper argues that a single RGB sequence, used to fit a Gaussian-splat 3D model while jointly optimizing per-part screw axes, is enough to reconstruct geometry, segment an object into rigid movable parts, and recover each part's joint axis. If this holds, a robot can get a usable kinematic model of a door, drawer, or laptop from ordinary video and can then manipulate the object from a text instruction. The practical stake is that the assumptions that normally limit articulated-object systems—extra sensors, known part count, and brittle intermediate estimates—can be dropped.

What carries the argument

The central object is the screw axis, the line (with an associated pitch) about which a rigid body moves; for a typical door or laptop hinge the pitch is zero and the axis is simply the hinge line. The paper couples each part's motion to a screw axis through the exponential map of rigid-body motion, and it renders the scene with Gaussian Splatting, whose Gaussian parameters define the 3D geometry. The mechanism doing the work is joint optimization: screw axes are initialized randomly and then updated, along with the splat parameters and part segmentation, so that photorealistic rendered views align with the input RGB frames. That alignment is what lets appearance alone determine both shape and kinematics.

What would settle it

Run the method on a video of a rigid object that never moves: a correct method should return one part and no meaningful screw axis, while a failure mode would imagine spurious joints from lighting or camera motion. A second test is a textureless moving door: if the hinge axis is not recovered because appearance provides no trackable features, the appearance-only claim is refuted.

Watch

Extended reading notes

Core claim

On its own terms, ScrewSplat's discovery is that screw axes can be made the carrier of articulation recovery inside a differentiable rendering loop. Starting from randomly initialized screw axes, the method iteratively optimizes them together with the parameters of several Gaussian-splat part models so that the rendered images match the observed RGB frames. The recovered per-part screw axes double as the kinematic model, and the splat parts double as the rigid-body segmentation, so recognition and reconstruction emerge from one optimization rather than from separate modules. The paper reports that this single end-to-end pipeline achieves the best recognition accuracy among the methods it compares against, and that the recovered model can be used for zero-shot manipulation guided by text.

Load-bearing premise

The load-bearing premise is that articulation is identifiable from appearance alone: each movable part must actually move during the input sequence, its surface must be textured or lit enough for photorealism to pin down that motion, and the non-convex joint optimization over splat parameters and screw axes, started from random axes, must land in the right basin.

Editorial extensions

If this is right

  • A robot could acquire a manipulable kinematic model of an articulated object from a single monocular video, with no depth sensor required.
  • Objects with an unknown number of moving parts can in principle be handled, because the part structure is a product of optimization rather than a preset count.
  • The recovered screw axes provide a grounding between language and motion, which the paper uses to enable zero-shot text-guided manipulation.
  • Eliminating intermediate stages such as part-count estimation and separate pose or joint estimation removes a source of accumulated error that prior pipelines suffer from.

Reading between the lines

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

  • The same optimization should be testable on objects whose parts never move during the video: if no joint motion is visible to constrain the axis, the method should either fail to detect a joint or recover an arbitrary axis, which would mark the boundary of the method's applicability.
  • Because the only supervision is photorealism, textureless or specular parts may under-constrain the screw axis; this is a direct, testable consequence of the appearance-only assumption.
  • A natural extension is to apply the recovered kinematic model as a prior for novel-view prediction or planning: once the screw axes are known, new poses of the object can be rendered without re-optimization.
  • The approach can be seen as folding the classic problem of inferring joint axes from relative motion directly into the reconstruction itself, which could also be applied to camera-rig calibration or hand-eye calibration settings.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes ScrewSplat, an end-to-end method for articulated object recognition from RGB observations. The method randomly initializes screw axes, jointly optimizes them with a Gaussian Splatting scene representation to reconstruct geometry and segment movable parts, and uses the recovered kinematic model for zero-shot text-guided manipulation. The abstract claims state-of-the-art recognition accuracy across diverse articulated objects. The full text as provided is severely corrupted: equations, tables, and most narrative passages are unreadable or garbled, so the derivation, optimization details, dataset splits, and numerical results cannot be verified from this submission.

Significance. If the claims hold, the contribution is significant: the method removes depth inputs, known part counts, and two-stage pipelines, and it couples 3D reconstruction with articulation discovery in a single photometric optimization. The screw-axis parameterization is a natural fit for kinematic structure, and the zero-shot text-guided manipulation result would be practically valuable. The conceptual simplicity is a real strength, and the claimed evaluation spans multiple articulated-object categories. However, no code, no convergence analysis, and no identifiability analysis are visible, and the empirical evidence is presented in tables that cannot be read in this submission. The significance therefore rests almost entirely on unverifiable SOTA claims, which is not sufficient for a journal decision.

major comments (3)
  1. [Abstract and Section 1] The central claim — that RGB-only joint optimization over randomly initialized screw axes recovers true articulation — needs identifiability and nonconvexity evidence. The visible text does not state whether every part is guaranteed to move in the input sequence, does not distinguish the axis-location ambiguity for prismatic joints, and does not explain why redundant screw axes cannot overfit the photometric loss. Please add a static-part control, a prismatic-joint control, a part-count over/under-specification sweep, and seed-variance statistics; without these, the SOTA tables alone cannot establish that the recovered axes are the true ones.
  2. [Section 4 (evaluation tables)] The reported recognition-accuracy and manipulation-success tables are not legible in this submission, and no confidence intervals, seed variance, per-part-type breakdowns, or axis-error metrics can be discerned. Because the method is stochastic in the screw-axis initialization and the abstract claims state-of-the-art performance, the paper must report distributions over restarts and ablate the same method with known part counts and with depth-based variants. A readable version of the evaluation tables with these statistics is required.
  3. [Section 3 (method and loss)] The loss functions, regularization terms, initialization scheme, and the exact optimization procedure are central to the paper's claims, but the equations in the submitted PDF are corrupted and cannot be checked. The photometric objective, the treatment of part counts, and the random screw-axis initialization should be restated in a readable form. This is a presentation blocker rather than a scientific flaw, but it makes verification impossible and must be fixed before the paper can be evaluated.
minor comments (4)
  1. [Abstract] The 'state-of-the-art' claim in the abstract should either specify the benchmark and metric or be softened until the evaluation tables can be verified.
  2. [Section 3] The phrase 'randomly initializing screw axes' should clarify what distribution is used and whether multiple restarts are performed; otherwise the reader cannot assess the role of initialization in the optimization.
  3. [References and reproducibility] The project website link is non-archival; please include dataset splits, hyperparameters, and evaluation protocols in the paper itself so the experiments are reproducible.
  4. [Conclusion / Limitations] The garbled text near the end appears to contain a limitations discussion; if the paper acknowledges unidentifiability for static parts or other failure modes, those caveats should be placed prominently in the method section and reflected in the abstract's scope.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ScrewSplat's core claim is an empirical optimization pipeline evaluated against external benchmarks with no fitted input relabeled as prediction.

full rationale

Walked the derivation chain visible in the provided text. The method takes RGB observations, randomly initializes screw axes, and jointly optimizes them with Gaussian-splat parameters against a photometric rendering loss; the recovered kinematic model is then evaluated against external articulated-object benchmarks and used for a downstream manipulation task. I found no step in which an output quantity is defined in terms of the claimed prediction, no fitted parameter relabeled as a prediction, and no load-bearing self-citation or imported uniqueness theorem. The abstract's statement that the approach 'begins by randomly initializing screw axes, which are then iteratively optimized' describes an optimization procedure rather than a tautology; whether the optimization is identifiable and avoids poor local minima is a correctness and convergence concern, not a circularity concern. The heavily garbled full text prevented reading the detailed equations and reference list, but nothing legible exhibits Equation X equaling Equation Y by construction or a self-citation standing in for an unverified premise. Under the hard rule that circularity requires a quotable reduction, the honest finding is no significant circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

From the abstract and legible fragments, the method reuses standard kinematic and rendering machinery (screw theory, Gaussian Splatting) rather than postulating new physical entities, so the invented-entities ledger is empty. The free parameters listed are the usual optimization choices that the abstract does not specify, and each would need to be checked against the full text to assess its effect on the central claim.

free parameters (3)
  • number of screw axes (movable parts)
    The abstract claims the method avoids a known part count, so the number of screw axes must be inferred or thresholded somewhere in the pipeline; the selection rule is not visible in the abstract.
  • loss weights for photometric and regularization terms
    A multi-objective optimization of rendering fidelity and kinematic priors typically requires manually chosen weights; the values are not stated in the abstract.
  • Gaussian Splatting hyperparameters
    Optimizer schedules, densification thresholds, and splat variance settings would need to be chosen; none are specified in the abstract.
assumptions (4)
  • standard math Rigid motion: each movable part moves as a rigid body about a fixed screw axis (Chasles' theorem).
    The entire method is built on the screw-axis representation of revolute and prismatic joints; this is a classical kinematics result implied by the title and abstract.
  • domain assumption Photometric consistency: matching rendered splat images to RGB observations is sufficient to identify geometry and joint axes.
    The end-to-end optimization assumes that appearance reconstruction error drives correct kinematic segmentation; this is the identifiable-from-appearance premise and is not proven in the abstract.
  • domain assumption Gaussian Splatting provides a differentiable renderer with adequate gradients for screw-axis optimization.
    The method depends on the smoothness and differentiability properties of the splatting renderer, which are empirical properties of the implementation rather than a theorem.
  • domain assumption The input RGB sequence shows the object in multiple articulation configurations.
    Without observed motion of each part, screw axes cannot be identified; the abstract does not discuss minimum sequence or motion requirements.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ScrewSplat: An End-to-End Method for Articulated Object Recognition." pith.science (2026). https://pith.science/paper/5T6QDCEP

@misc{pith2026250802146,
  author       = {Pith},
  title        = {Pith review of: ScrewSplat: An End-to-End Method for Articulated Object Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5T6QDCEP}},
  note         = {Machine review of arXiv:2508.02146}
}
read the original abstract

Articulated object recognition -- the task of identifying both the geometry and kinematic joints of objects with movable parts -- is essential for enabling robots to interact with everyday objects such as doors and laptops. However, existing approaches often rely on strong assumptions, such as a known number of articulated parts; require additional inputs, such as depth images; or involve complex intermediate steps that can introduce potential errors -- limiting their practicality in real-world settings. In this paper, we introduce ScrewSplat, a simple end-to-end method that operates solely on RGB observations. Our approach begins by randomly initializing screw axes, which are then iteratively optimized to recover the object's underlying kinematic structure. By integrating with Gaussian Splatting, we simultaneously reconstruct the 3D geometry and segment the object into rigid, movable parts. We demonstrate that our method achieves state-of-the-art recognition accuracy across a diverse set of articulated objects, and further enables zero-shot, text-guided manipulation using the recovered kinematic model. See the project website at: https://screwsplat.github.io.

Discussion (0). Sign in to comment.

Forward citations

Cited by 3 Pith papers

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

  1. StructureGS: Structure-aware Gaussian Splatting for Articulated Object Reconstruction

    cs.GR 2026-07 conditional novelty 6.0 of 10

    OBB-based part-fitting and contact losses on 3D Gaussians disentangle geometry, appearance, and motion for cleaner articulated reconstruction than photometric-only baselines.

  2. PokeNet: Learning Kinematic Models of Articulated Objects from Human Observations

    cs.RO 2026-02 conditional novelty 6.0 of 10

    PokeNet estimates joint types, axes, ranges, and operation order of articulated objects directly from a single-view point cloud video of a human demonstration.

  3. Advances in 4D Representation: Geometry, Motion, and Interaction

    cs.CV 2025-10 conditional novelty 4.0 of 10

    A representation-centric survey of 4D generation and reconstruction, organized by geometry, motion, and interaction, with qualitative trade-off comparisons across seven representation families.

Reference graph

Works this paper leans on

51 extracted references · 42 canonical work pages · cited by 3 Pith papers

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    X. Li, H. Wang, L. Yi, L. J. Guibas, A. L. Abbott, and S. Song. Category-level articulated object pose estimation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3706--3715, 2020

  3. [3]

    V. Zeng, T. E. Lee, J. Liang, and O. Kroemer. Visual identification of articulated object parts. In 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 2443--2450. IEEE, 2021

  4. [4]

    A. Jain, R. Lioutikov, C. Chuck, and S. Niekum. Screwnet: Category-independent articulation model estimation from depth images using screw theory. In 2021 IEEE International Conference on Robotics and Automation (ICRA), pages 13670--13677. IEEE, 2021

  5. [5]

    J. Mu, W. Qiu, A. Kortylewski, A. Yuille, N. Vasconcelos, and X. Wang. A-sdf: Learning disentangled signed distance functions for articulated shape representation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 13001--13011, 2021

  6. [6]

    Tseng, H.-J

    W.-C. Tseng, H.-J. Liao, L. Yen-Chen, and M. Sun. Cla-nerf: Category-level articulated neural radiance field. In 2022 International Conference on Robotics and Automation (ICRA), pages 8454--8460. IEEE, 2022

  7. [7]

    A. Jain, S. Giguere, R. Lioutikov, and S. Niekum. Distributional depth-based estimation of object articulation models. In Conference on Robot Learning, pages 1611--1621. PMLR, 2022

  8. [8]

    F. Wei, R. Chabra, L. Ma, C. Lassner, M. Zollh \"o fer, S. Rusinkiewicz, C. Sweeney, R. Newcombe, and M. Slavcheva. Self-supervised neural articulated shape and appearance models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15816--15826, 2022

Show all 51 references
  1. [9]

    Jiang, C.-C

    Z. Jiang, C.-C. Hsu, and Y. Zhu. Ditto: Building digital twins of articulated objects from interaction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5616--5626, 2022

  2. [10]

    Heppert, T

    N. Heppert, T. Migimatsu, B. Yi, C. Chen, and J. Bohg. Category-independent articulated object tracking with factor graphs. In 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3800--3807. IEEE, 2022

  3. [11]

    Heppert, M

    N. Heppert, M. Z. Irshad, S. Zakharov, K. Liu, R. A. Ambrus, J. Bohg, A. Valada, and T. Kollar. Carto: Category and joint agnostic reconstruction of articulated objects. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21201--21210, 2023

  4. [12]

    J. Lei, C. Deng, W. B. Shen, L. J. Guibas, and K. Daniilidis. Nap: Neural 3d articulated object prior. Advances in Neural Information Processing Systems, 36: 0 31878--31894, 2023

  5. [13]

    Mildenhall, P

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, and R. Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. Communications of the ACM, 65 0 (1): 0 99--106, 2021

  6. [14]

    P. Wang, L. Liu, Y. Liu, C. Theobalt, T. Komura, and W. Wang. Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction. arXiv preprint arXiv:2106.10689, 2021

  7. [15]

    Kerbl, G

    B. Kerbl, G. Kopanas, T. Leimk \"u hler, and G. Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 42 0 (4): 0 139--1, 2023

  8. [16]

    B. Wen, J. Tremblay, V. Blukis, S. Tyree, T. M \"u ller, A. Evans, D. Fox, J. Kautz, and S. Birchfield. Bundlesdf: Neural 6-dof tracking and 3d reconstruction of unknown objects. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 606--617, 2023

  9. [17]

    J. Liu, A. Mahdavi-Amiri, and M. Savva. Paris: Part-level reconstruction and motion analysis for articulated objects. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 352--363, 2023

  10. [18]

    Y. Weng, B. Wen, J. Tremblay, V. Blukis, D. Fox, L. Guibas, and S. Birchfield. Neural implicit representation for building digital twins of unknown articulated objects. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3141--3150, 2024

  11. [19]

    J. Kerr, C. M. Kim, M. Wu, B. Yi, Q. Wang, K. Goldberg, and A. Kanazawa. Robot see robot do: Imitating articulated object manipulation with monocular 4d reconstruction. arXiv preprint arXiv:2409.18121, 2024

  12. [20]

    Y. Liu, B. Jia, R. Lu, J. Ni, S.-C. Zhu, and S. Huang. Building interactable replicas of complex articulated objects via gaussian splatting. arXiv preprint arXiv:2502.19459, 2025

  13. [21]

    J. Guo, Y. Xin, G. Liu, K. Xu, L. Liu, and R. Hu. Articulatedgs: Self-supervised digital twin modeling of articulated objects using 3d gaussian splatting. arXiv preprint arXiv:2503.08135, 2025

  14. [22]

    K. M. Lynch and F. C. Park. Modern robotics. Cambridge University Press, 2017

  15. [23]

    R. Liu, A. Canberk, S. Song, and C. Vondrick. Differentiable robot rendering. arXiv preprint arXiv:2410.13851, 2024

  16. [24]

    Radford, J

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pages 8748--8763. PmLR, 2021

  17. [25]

    R. Gal, O. Patashnik, H. Maron, A. H. Bermano, G. Chechik, and D. Cohen-Or. Stylegan-nada: Clip-guided domain adaptation of image generators. ACM Transactions on Graphics (TOG), 41 0 (4): 0 1--13, 2022

  18. [26]

    Xiang, Y

    F. Xiang, Y. Qin, K. Mo, Y. Xia, H. Zhu, F. Liu, M. Liu, H. Jiang, Y. Yuan, H. Wang, et al. Sapien: A simulated part-based interactive environment. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11097--11107, 2020

  19. [27]

    B. O. Community. Blender - a 3D modelling and rendering package. Blender Foundation, Stichting Blender Foundation, Amsterdam, 2018. URL http://www.blender.org

  20. [28]

    Bolanos, S.-Y

    L. Bolanos, S.-Y. Su, and H. Rhodin. Gaussian shadow casting for neural characters. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20997--21006, 2024

  21. [29]

    Loper, N

    M. Loper, N. Mahmood, J. Romero, G. Pons-Moll, and M. J. Black. Smpl: A skinned multi-person linear model. In Seminal Graphics Papers: Pushing the Boundaries, Volume 2, pages 851--866. 2023

  22. [30]

    J. Lin, L. Zhang, K. Lee, J. Ning, J. Goldfeder, and H. Lipson. Autourdf: Unsupervised robot modeling from point cloud frames using cluster registration. arXiv preprint arXiv:2412.05507, 2024

  23. [31]

    Viswanath, K

    V. Viswanath, K. Shivakumar, M. Parulekar, J. Ajmera, J. Kerr, J. Ichnowski, R. Cheng, T. Kollar, and K. Goldberg. Handloom: Learned tracing of one-dimensional objects for inspection and manipulation. In Conference on Robot Learning, pages 341--357. PMLR, 2023

  24. [32]

    C. Chi, B. Burchfiel, E. Cousineau, S. Feng, and S. Song. Iterative residual policy: for goal-conditioned dynamic manipulation of deformable objects. The International Journal of Robotics Research, 43 0 (4): 0 389--404, 2024

  25. [33]

    Varley, C

    J. Varley, C. DeChant, A. Richardson, J. Ruales, and P. Allen. Shape completion enabled robotic grasping. In 2017 IEEE/RSJ international conference on intelligent robots and systems (IROS), pages 2442--2447. IEEE, 2017

  26. [34]

    H. Kato, Y. Ushiku, and T. Harada. Neural 3d mesh renderer. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 3907--3916, 2018

  27. [35]

    S. Liu, S. Saito, W. Chen, and H. Li. Learning to infer implicit surfaces without 3d supervision. Advances in Neural Information Processing Systems, 32, 2019

  28. [36]

    J. J. Park, P. Florence, J. Straub, R. Newcombe, and S. Lovegrove. Deepsdf: Learning continuous signed distance functions for shape representation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 165--174, 2019

  29. [37]

    Van der Merwe, Q

    M. Van der Merwe, Q. Lu, B. Sundaralingam, M. Matak, and T. Hermans. Learning continuous 3d reconstructions for geometrically aware grasping. In 2020 IEEE International Conference on Robotics and Automation (ICRA), pages 11516--11522. IEEE, 2020

  30. [38]

    Ichnowski, Y

    J. Ichnowski, Y. Avigal, J. Kerr, and K. Goldberg. Dex-nerf: Using a neural radiance field to grasp transparent objects. arXiv preprint arXiv:2110.14217, 2021

  31. [39]

    J. Kerr, L. Fu, H. Huang, Y. Avigal, M. Tancik, J. Ichnowski, A. Kanazawa, and K. Goldberg. Evo-nerf: Evolving nerf for sequential robot grasping of transparent objects. In 6th annual conference on robot learning, 2022

  32. [40]

    S. Kim, T. Ahn, Y. Lee, J. Kim, M. Y. Wang, and F. C. Park. Dsqnet: A deformable model-based supervised learning algorithm for grasping unknown occluded objects. IEEE Transactions on Automation Science and Engineering, 20 0 (3): 0 1721--1734, 2022

  33. [41]

    Q. Dai, Y. Zhu, Y. Geng, C. Ruan, J. Zhang, and H. Wang. Graspnerf: Multiview-based 6-dof grasp detection for transparent and specular objects using generalizable nerf. arXiv preprint arXiv:2210.06575, 2022

  34. [42]

    S. Kim, Y. H. Kim, Y. Lee, and F. C. Park. Leveraging 3d reconstruction for mechanical search on cluttered shelves. In 7th Annual Conference on Robot Learning, 2023

  35. [43]

    Y. H. Kim, S. Kim, Y. Lee, and F. C. Park. T ^ 2 sqnet: A recognition model for manipulating partially observed transparent tableware objects. In 8th Annual Conference on Robot Learning, 2024

  36. [44]

    Y. H. Kim, S. Kim, Y. Lee, and F. C. Park. Dreamgrasp: Zero-shot 3d multi-object reconstruction from partial-view images for robotic manipulation. arXiv preprint arXiv:2507.05627, 2025

  37. [45]

    Abbatematteo, S

    B. Abbatematteo, S. Tellex, and G. Konidaris. Learning to generalize kinematic models to novel objects. In Proceedings of the 3rd Conference on Robot Learning, 2019

  38. [46]

    L. Le, J. Xie, W. Liang, H.-J. Wang, Y. Yang, Y. J. Ma, K. Vedder, A. Krishna, D. Jayaraman, and E. Eaton. Articulate-anything: Automatic modeling of articulated objects via a vision-language foundation model. arXiv preprint arXiv:2410.13882, 2024

  39. [47]

    S. Y. Gadre, K. Ehsani, and S. Song. Act the part: Learning interaction strategies for articulated object part discovery. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 15752--15761, 2021

  40. [48]

    C.-C. Hsu, Z. Jiang, and Y. Zhu. Ditto in the house: Building articulation models of indoor scenes through interactive perception. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 3933--3939. IEEE, 2023

  41. [49]

    L. Ma, J. Meng, S. Liu, W. Chen, J. Xu, and R. Chen. Sim2real 2: Actively building explicit physics model for precise articulated object manipulation. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 11698--11704. IEEE, 2023

  42. [50]

    W. E. Lorensen and H. E. Cline. Marching cubes: A high resolution 3d surface construction algorithm. In Seminal graphics: pioneering efforts that shaped the field, pages 347--353. 1998

  43. [51]

    Kirillov, E

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y. Lo, P. Doll \'a r, and R. Girshick. Segment anything. arXiv:2304.02643, 2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.