Pith. sign in

REVIEW 5 major objections 5 minor 2 cited by

DexVLG: Dexterous Vision-Language-Grasp Model at Scale

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Language-guided robot hand executes zero-shot grasps 76% of the time

desk verdict A large, genuinely useful grasp dataset and a plausible model, but the evaluation doesn't yet prove the language-conditioning claim. read the letter →

arxiv 2507.02747 v1 pith:LNAQVCJA submitted 2025-07-03 cs.CV cs.RO

Jiawei He , Danshi Li , Xinqiang Yu , Zekun Qi , Wenyao Zhang , Jiayi Chen , Zhaoxiang Zhang , Zhizheng Zhang
show 2 more authors
Li Yi He Wang
This is my paper · ORCID
classification cs.CVcs.RO
keywords dexterousgraspingvision-language-actionmodellanguage-conditionedpart-awaregraspsynthesisflowmatchingDexNet3.0zero-shotgeneralizationRGBDpointcloud
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

This paper tries to establish that language-conditioned dexterous grasping can be scaled to thousands of object categories by training a large vision-language-grasp model on a synthetic dataset, rather than collecting human demonstrations. The authors built DexGraspNet 3.0, 170 million part-aligned grasp poses on 174,000 objects, each validated in physics simulation and paired with a caption naming the part and contact fingers, and used it to train DexVLG, which takes a single-view colored point cloud and an instruction like "grasp the handle of the hammer" and outputs a Shadow Hand pose. They report over 76% zero-shot execution success in simulation, state-of-the-art part-grasp accuracy, and 80% real-world success on simple objects. If the paper is right, functional, instruction-following dexterous grasping becomes a data-and-model-scaling problem rather than a teleoperation bottleneck.

What carries the argument

The load-bearing machinery is the DexGraspNet 3.0 synthesis pipeline plus the DexVLG architecture. The pipeline segments each object mesh into parts, captions parts automatically, initializes hand poses from oriented part bounding boxes, and optimizes grasps with a truncated-barrier part-contact energy that repels fingertips from surfaces outside the target part, followed by physics validation including a part-alignment condition. The model couples a Uni3D point-cloud encoder with a Florence-2 backbone and a flow-matching denoising head that shares transformer layers with the language model, so the generated grasp is conditioned directly on language tokens.

What would settle it

Re-annotate a sample of objects with human part labels and re-run DexVLG's part-touch and part-grasp accuracy using those labels instead of the automatic labels; if the scores fall to chance while the original metrics stay high, the alignment metrics were measuring label consistency, not genuine language-aligned grasping.

Watch

Extended reading notes

Core claim

The central claim is that language-aligned dexterous grasp pose prediction is achievable at scale: a flow-matching pose head conditioned on a fine-tuned Florence-2 language model and a Uni3D point-cloud encoder can generate grasps that touch the requested semantic part, generalize to objects unseen during training, and transfer to physical execution. The authors support this with DexGraspNet 3.0, a dataset of 170M validated grasps across 174k objects, each pose passing a simulation-based part-alignment check and paired with a caption; on their simulation benchmarks DexVLG outperforms retrained small-model baselines with and without CLIP text conditioning, reaching over 76% execution success and the highest part-grasp accuracy.

Load-bearing premise

The load-bearing premise is that the automatic part segmentation and captioning produces semantically correct, human-compatible part labels; if a label only describes a geometric chunk, both the training signal and the part-accuracy metrics are biased in the same direction.

Editorial extensions

If this is right

  • Language-conditioned dexterous grasping can be trained end-to-end from synthetic data at scale, without human teleoperation, and generalizes zero-shot to unseen objects in simulation.
  • The DexGraspNet 3.0 dataset provides a reusable 170M-pose resource for training and benchmarking future vision-language-action models that control multi-fingered hands.
  • Part-contact barrier energy plus simulation validation is enough to make grasps land on the requested part, enabling functional grasps like holding a hammer by the handle.
  • Color is a decisive input signal: removing it drops simulation success by up to 27.7 points, so single-view RGBD is the right modality for part-level grasping.
  • Large model capacity is necessary for instruction-aligned dexterous grasping; adding CLIP text conditioning to a small diffusion model trades away grasp quality without matching part accuracy.

Reading between the lines

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

  • Because the part labels and the evaluation metric come from the same pipeline, the reported part accuracy may overstate true language understanding; a human-annotated part-labels benchmark would settle how much of the gain is semantic versus label-geometry consistency.
  • The 80% real-world success on simple objects suggests the main barrier to harder objects is the hand-arm workspace constraint the authors identify, so filtering training data for arm-reachable poses could extend the result.
  • The flow-matching-plus-LLM conditioning design is not hand-specific and could be reused for other high-dimensional action spaces, such as bimanual hands or whole-arm grasps.
  • Since the paper reports no ranking of generated samples, deployment would need a separate grasp scorer; the authors note that likelihood-based ranking is infeasible with VLM gradients.
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

5 major / 5 minor

Summary. The paper presents DexGraspNet 3.0, a synthetic dataset of 170 million dexterous grasp poses over 174k Objaverse objects, with SAMesh part segmentation and GPT-4o-generated part captions, and DexVLG, a vision-language-grasp model that combines Uni3D, Florence-2, and a flow-matching pose head to predict language-conditioned dexterous grasps from single-view RGBD point clouds. The authors report over 76% zero-shot simulation success and state-of-the-art part-grasp accuracy on curated benchmarks, an ablation study on data scale, model size, and color input, and an 80% real-world success rate on simple objects. The paper is clearly written and the dataset construction pipeline is described in unusual detail in the appendix.

Significance. If the central claims hold, this is a substantial contribution: DexGraspNet 3.0 would be the largest dexterous grasp dataset with semantic part-level captions by a wide margin, and DexVLG would demonstrate a practical path toward language-conditioned functional dexterous grasping. The authors deserve credit for releasing a project page, documenting the simulation validation criteria, and including ablations for data scale, model size, and input modality. However, the most important evaluative claims—namely that the model's grasps are language-aligned and generalize to unseen objects—are currently supported only by metrics that are partly self-referential and by benchmarks that exclude difficult cases. The paper is therefore significant but needs additional controls and transparency before the headline claims can be accepted.

major comments (5)
  1. [§6.2, §6.3, §4.4] The part-alignment metrics PTA and PGA are computed against the same SAMesh/GPT-4o part labels that were used to build the DexGraspNet 3.0 training captions, and the metric's 'closer to the desired part than any other part' condition directly mirrors dataset-validation criterion 4 in §4.4. This shared annotation source and shared geometric condition make the reported part accuracy a measure of self-consistency with the automatic pipeline rather than of human-compatible functional alignment. The authors should report PTA/PGA on a human-verified subset of part labels, or use an independently segmented and annotated evaluation set, and should explicitly quantify the label agreement between SAMesh/GPT-4o and human judgment.
  2. [§6.3.1, Table 4] No experiment isolates the language instruction. The no-instruction row in Table 4 reports only Suc, with PTA/PGA omitted ('–'), and every benchmark instruction in §6.3 names a part that exists and is graspable on the object. A model that emits a generic functional grasp would receive substantial PTA/PGA whenever the requested part happens to be the one it touches. The authors should add an instruction-swap control—the same object and point cloud with different part instructions—and report PTA/PGA for the no-instruction model. Without this, the central claim that DexVLG's grasps are language-aligned is not established.
  3. [§6.3] The simulation benchmarks are explicitly curated: 'we manually select well-segmented objects and filter their poses on table such that the target parts for grasping are not occluded by the table.' This exclusion removes precisely the cases where automatic part segmentation or table occlusion would stress the model, and it weakens the zero-shot generalization claim. The authors should report how many objects and poses were excluded, and provide results on an unfiltered or randomly sampled subset, or justify that the filtering does not bias the comparison in favor of their method.
  4. [§6.4] The real-world experiment reports only '80% success rate and 75% part accuracy with these simple objects' and gives no trial counts, no object list, no definition of the human part-accuracy check, and no selection criteria for the objects. The real-world claim is currently anecdotal; the authors should report per-object trial counts, confidence intervals, and a detailed protocol for the human part-accuracy assessment.
  5. [Table 6, §6.5] The model-size ablation is internally inconsistent with the main results. Table 6 reports that the 1B model achieves LVIS-Seen Suc 75.3 and PGA 39.1, whereas Table 2 reports DexVLG—which is the same 1B configuration per §5—at 87.7 Suc and 62.1 PGA. The numbers in the Table 6 1B row are identical to the 'PC w/ color' row in Table 7. The reader cannot tell which setting Table 6 describes, so the conclusions drawn from it about model size are not currently supported and should be corrected or clarified.
minor comments (5)
  1. [§6.3] The text refers to 'DexGraspNet2.0 [56]' but reference [56] is the original DexGraspNet paper; the DexGraspNet 2.0 citation appears to be [67]. Please verify all such citations.
  2. [Table 6] The caption says 'The 225M model' but the table row lists 255M; the parameter counts should be checked for consistency.
  3. [Table 4] The no-instruction row should report PTA/PGA if these metrics were computed; if they were not, the text should state explicitly that language alignment was not measured in this condition.
  4. [§5.3, Eq. (5)] The flow-matching objective does not explicitly show the conditioning on the LLM hidden states; adding the conditioning variable to v would make the architecture description clearer.
  5. [§4.3.2, Eq. (1)] In the part-contact energy, the index j is used for 'object surface outside the target part' but the target part index i is not formally introduced; please define i in the equation or its preceding sentence.

Circularity Check

2 steps flagged · score 4.0 of 10

PTA/PGA part-alignment metrics restate the dataset's own part-contact filter and reuse the same SAMesh/GPT-4o labels as training, making the headline part-accuracy numbers partly self-referential; SamPart3D and human real-world checks keep the central claim partially independent.

  1. self definitional [Sec. 4.3.2 and Sec. 4.4 (dataset construction) vs Sec. 6.2 (PTA/PGA metrics)]
    "Hence strictly enforces part alignment when the stepsize is small enough. ... 4. the part-alignment condition, which indicates that if a hand link is in contact with the object (i.e. the distance to the object is less than 0.2 cm), this hand link should be closer to the desired part than any other parts of the object. ... Part Touch Accuracy (PTA) represents the percentage of grasp poses that touch the target semantic part, evaluated by checking whether at least one finger in the predicted pose is less than 1cm away from the part and is closer to the desired part than any other part."

    The training set is produced by an optimizer that 'strictly enforces part alignment' via E_bar, and the dataset acceptance test adds the same closest-part rule as criterion 4. PTA (and PGA, with three fingers) then score the model by that same closest-part rule, relaxed from 0.2 cm to 1 cm. Thus the headline part-accuracy metric is a restatement of the data-generation filter: it rewards the model for reproducing the geometric condition that every training grasp already satisfies, and it does not test whether the SAMesh/GPT-4o part label or the grasp is semantically correct for a human. The physics-based Suc metric is independent, but PTA/PGA are not an external check on language-aligned functional grasping.

  2. self definitional [Sec. 4.2 (part labeling) vs Sec. 6.3 (LVIS-Seen/Unseen benchmarks) and Sec. 6.2 (PTA/PGA ground truth)]
    "We use SAMesh [47] to perform zero-shot geometry-based part segmentation on colorless collision meshes. ... part names are automatically labeled using set-of-mark prompting [63] with GPT-4o. ... The LVIS-Seen benchmark consists of 40 seen objects in the Objaverse-LVIS split. ... The Unseen benchmark consists of 84 Objaverse objects unseen in the training process. ... The SamPart3D benchmark consists of 56 Objaverse objects segmented and semantically annotated by SamPart3D [65] using methods different from our work."

    Part names used as training captions and as the 'target semantic part' in PTA/PGA come from the same automatic pipeline (SAMesh geometry segmentation plus GPT-4o set-of-mark labeling). For LVIS-Seen and Unseen, the paper introduces no alternative label source; only the SamPart3D benchmark is explicitly said to use 'methods different from our work'. Consequently, on the two primary benchmarks, any systematic error in SAMesh/GPT-4o part semantics is shared by the training target and the evaluation ground truth, so high PTA/PGA can reflect self-consistency with the labeler rather than human-meaningful part alignment. The SamPart3D benchmark and human-checked real-world part accuracy provide partial external grounding, which limits the circularity.

full rationale

Score 4 reflects one genuinely self-referential link in the evaluation chain. The dataset is filtered by a 'part-alignment condition' (closest-part contact) and by an E_bar objective that 'strictly enforces part alignment'; PTA/PGA then measure exactly that same closest-part condition, so the part-accuracy headline is partly a measure of how well the model imitates the training filter. On LVIS-Seen and Unseen, the target parts are labeled by the same SAMesh/GPT-4o pipeline that produced training captions, so the metric cannot detect a systematic semantic bias in those labels. This is not a fully circular derivation: the model must still generate physically valid, part-specific poses and Suc is an independent physics check; the SamPart3D benchmark uses a different segmenter (66.0 PTA / 52.0 PGA) and real-world part accuracy is human-examined (75%), giving partial external grounding. The paper's self-citations to DexGraspNet 1.0/2.0 are ordinary method reuse and are not load-bearing; no uniqueness theorem is imported. The missing no-instruction PTA/PGA control is a confound that weakens the language-alignment conclusion but is not itself a circular step.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central claims rest on an elaborate synthetic pipeline. The most important unpaid inputs are physical simulation fidelity, semantic validity of SAMesh and GPT-4o labels, manual benchmark curation, and a long list of hand-chosen thresholds. The paper introduces no new physical entities; its new artifacts are the dataset and the model, both described in the manuscript.

free parameters (5)
  • Energy weights omega_FC, omega_bar, omega_dis, omega_reg = not reported
    Equation 4 balances force closure, part barrier, distance, and regularization energies in grasp synthesis; the reported results depend on these values, but the values are not given.
  • Distance thresholds d0, d_thr = d0=1cm; d_thr not reported
    Equations 1-3 set the palm-object distance and the truncation threshold for the part barrier; both are hand-chosen and affect which grasps are generated.
  • Simulation validation thresholds = penetration under 3mm, self-penetration under 3mm, part contact under 0.2cm
    Section 4.4 defines which poses enter the dataset and what PTA and PGA later measure, so these thresholds directly shape the central claims.
  • Object size normalization range = [20,50] cm diagonal
    Appendix B clips object sizes to a graspable range, biasing the dataset and downstream evaluation to a particular scale.
  • Initialization budget and hand-crafted joint poses = 5000 poses per part per split; jitter angles not fully specified
    Appendix C uses four geometric categories and hand-crafted initial joint and contact settings; these choices determine grasp diversity and quality.
assumptions (6)
  • domain assumption IsaacGym physics simulation predicts real-world grasp success.
    Training labels, dataset validation, and evaluation success all use IsaacGym; sim-to-real transfer is assumed rather than demonstrated at scale.
  • domain assumption SAMesh zero-shot part segmentation yields consistent semantic parts across objects.
    Section 4.2 uses SAMesh to define parts; both dataset construction and the main benchmarks inherit this segmentation.
  • domain assumption GPT-4o produces accurate object and part names and size estimates from rendered views.
    Section 4.2 and Appendix B rely on GPT-4o for filtering, captioning, and size estimation, all of which determine dataset content.
  • domain assumption Manually curated benchmarks are representative of tabletop dexterous grasping.
    Section 6.3 selects well-segmented objects and filters out occluded poses; the representativeness claim is unverified.
  • domain assumption Force closure as approximated by DFC and LP-based DFC correlates with real grasp stability.
    Section A.1-A.3: the objective optimized to synthesize all dataset grasps assumes force closure is a sufficient guide to stable, natural grasps.
  • standard math Flow-matching objective learns the conditional distribution of valid grasps.
    Section 5.3 Equation 5 applies the standard flow-matching assumption that minimizing the velocity objective yields a usable generative model.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DexVLG: Dexterous Vision-Language-Grasp Model at Scale." pith.science (2026). https://pith.science/paper/LNAQVCJA

@misc{pith2026250702747,
  author       = {Pith},
  title        = {Pith review of: DexVLG: Dexterous Vision-Language-Grasp Model at Scale},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LNAQVCJA}},
  note         = {Machine review of arXiv:2507.02747}
}
read the original abstract

As large models gain traction, vision-language-action (VLA) systems are enabling robots to tackle increasingly complex tasks. However, limited by the difficulty of data collection, progress has mainly focused on controlling simple gripper end-effectors. There is little research on functional grasping with large models for human-like dexterous hands. In this paper, we introduce DexVLG, a large Vision-Language-Grasp model for Dexterous grasp pose prediction aligned with language instructions using single-view RGBD input. To accomplish this, we generate a dataset of 170 million dexterous grasp poses mapped to semantic parts across 174,000 objects in simulation, paired with detailed part-level captions. This large-scale dataset, named DexGraspNet 3.0, is used to train a VLM and flow-matching-based pose head capable of producing instruction-aligned grasp poses for tabletop objects. To assess DexVLG's performance, we create benchmarks in physics-based simulations and conduct real-world experiments. Extensive testing demonstrates DexVLG's strong zero-shot generalization capabilities-achieving over 76% zero-shot execution success rate and state-of-the-art part-grasp accuracy in simulation-and successful part-aligned grasps on physical objects in real-world scenarios.

Figures

Figures reproduced from arXiv: 2507.02747 by the authors.

Figure 1
Figure 1. Overview. Our main contributions are twofold: First, we constructed a large-scale synthetic dexterous grasping dataset called DexGraspNet3.0, which contains grasp poses with captions describing the grasped part and style. Second, we trained a language-instructed grasp pose prediction model using the DexGraspNet3.0 dataset, called DexVLG. This model can generate language-aligned and generaliz￾able grasping poses for … view at source ↗
Figure 2
Figure 2. Visualization of part-aware dexterous grasp poses in DexGraspNet3.0. The left columns visualize sample objects together with part segmentation generated by SAMesh [47] and captioned by GPT-4o [39]. On the right are part-aligned grasp poses generated by our optimization pipeline. Each grasp makes contact with a single object part and naturally aligns with the way humans grasp objects [PITH_FULL_IMAGE:figures/full_fi… view at source ↗
Figure 3
Figure 3. Grasp pose generation pipeline. Given a watertight colorless mesh of an object, we perform part segmentation with SAMesh [47] and fit oriented tight bounding boxes for each part. On each part, we sample grasp points on certain areas and align hand initialization poses with the part by leveraging geometric cues in bounding box parameters and grasp points. We further jitter each initial hand pose and run batched gradi… view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: Pipeline of our DexVLG model. It first uses Uni3D [71] to encode the colored point cloud. Then, Florence-2 [60] is leveraged to reason the projected visual features and the language tokens. Next, a flow-matching denoise head will conditioned on the LLM output embedding…
Figure 5
Figure 5. Figure 5: Visualization of grasp poses predicted by DexVLG in simulation. The object mesh is drawn only for visualization. The model input is a single-view point cloud, the color present in this figure is painted only for visualization. Each grasp is instructed with ”Grasp the […
Figure 6
Figure 6. Figure 6: real-world experiment setting. the {object} object with contact on {fingers}”, where {fingers} are the names of fingers that we want to make contact with the object. We compare models trained with (dubbed “+contact label”) and without contact mode la￾bels on the LVIS-S…
Figure 7
Figure 7. Figure 7: Visualization of LP-based DFC vs DFC. Left: DFC generates artifacts such as tilted finger and drifted contacts, high￾lighted in red. Right: LP-based DFC generates more natural poses that are compliant with the geometry of the object, highlighted in green. The grasp pos…
Figure 8
Figure 8. Figure 8: Visualization of part segmentation and GPT generated part annotations [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: Visualization of initial hand joint positions and con￾tact candidates. Split Name Object Grasps Captions Contact Candidates Ours-Wrap 169k 103M 103M 7 Ours-Pinch 139k 67M 67M 4 Total 174k 170M 170M - [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 10
Figure 10. Figure 10: Lid-like part For lid-like parts, we identify the part’s principal direc- [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]
Figure 12
Figure 12. Figure 12: L-shaped part Shaft-like part is typical for handles of tools and con￾necting bars of long objects such as lamps and barbells. Ob￾ject parts that are not categorized into any of other types are categorized as shaft-like parts by default, because we em￾perically find t…
Figure 11
Figure 11. Figure 11: Disk-like part For disk-like parts, we identify the shortest axis of its bounding box that points into an adjacent part as the part principal direction. We sample grasp points on the surface area where the surface normal is perpendicular to the part principal directio…
Figure 14
Figure 14. Figure 14: Visualization of camera poses used to render the tabletop dataset. We place the object on the table with translation in the x and y direction zeroed. The red dot represents the place where the object is placed. We sample camera poses evenly in the circle 80cm away fro…
Figure 15
Figure 15. Figure 15: Comparison between grasp poses synthesized with random initialization colored in green v.s. our part-aligned initial￾ization colored in red. The part-aligned initialization injects a strong prior about how a natural grasp is likely to be posed. The resulting optimized…
Figure 16
Figure 16. Figure 16: Prompt used to query for part names of objects with GPT-4o. We use Set-of-Marks [63] with images rendered from six views. [NAME] and [DESCRIPTION] are the captions given in the OpenShape [31] dataset [PITH_FULL_IMAGE:figures/full_fig_p020_16.png]
Figure 17
Figure 17. Figure 17: Prompt used to query for object sizes with GPT-4o.[NAME] and [DESCRIPTION] are the captions given in the Open￾Shape [31] dataset [PITH_FULL_IMAGE:figures/full_fig_p021_17.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. WristMimic: Full-Body Humanoid Control with Wrist-Guided Manipulation

    cs.RO 2026-07 conditional novelty 6.0 of 10

    WristMimic achieves comparable or superior object manipulation retargeting by supervising wrist kinematics while letting finger behavior emerge from object and contact dynamics.

  2. Being-H0: Vision-Language-Action Pretraining from Large-Scale Human Videos

    cs.CV 2025-07 conditional novelty 6.0 of 10

    A dexterous VLA pretrained on a 2.5M-instance human hand motion dataset transfers skills to a real robot hand, outperforming baselines in manipulation tasks.

Reference graph

Works this paper leans on

73 extracted references · 43 canonical work pages · cited by 2 Pith papers

  1. [1]

    Gpt-4 technical report.arXiv preprint arXiv:2303.08774,

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ah- mad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report.arXiv preprint arXiv:2303.08774,

    arXiv

  2. [2]

    Dexterous functional grasping

    Ananye Agarwal, Shagun Uppal, Kenneth Shaw, and Deepak Pathak. Dexterous functional grasping. In7th Annual Con- ference on Robot Learning, 2023. 3

    work page 2023

  3. [3]

    Kevin Black, Noah Brown, Danny Driess, Adnan Esmail, Michael Equi, Chelsea Finn, Niccolo Fusai, Lachy Groom, Karol Hausman, Brian Ichter, et al.π 0: A vision-language- action flow model for general robot control.arXiv preprint arXiv:2410.24164, 2024. 2, 6

    arXiv

  4. [4]

    Rt-2: Vision-language-action models transfer web knowledge to robotic control.arXiv preprint arXiv:2307.15818, 2023

    Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, Xi Chen, Krzysztof Choromanski, Tianli Ding, Danny Driess, Avinava Dubey, Chelsea Finn, et al. Rt-2: Vision-language-action models transfer web knowledge to robotic control.arXiv preprint arXiv:2307.15818, 2023. 3

    arXiv 2023

  5. [5]

    Task-oriented dexterous grasp synthesis via differ- entiable grasp wrench boundary estimator.arXiv preprint arXiv:2309.13586, 2023

    Jiayi Chen, Yuxing Chen, Jialiang Zhang, and He Wang. Task-oriented dexterous grasp synthesis via differ- entiable grasp wrench boundary estimator.arXiv preprint arXiv:2309.13586, 2023. 2, 3

    arXiv 2023

  6. [6]

    Bodex: Scalable and efficient robotic dexterous grasp synthesis using bilevel op- timization.arXiv preprint arXiv:2412.16490, 2024

    Jiayi Chen, Yubin Ke, and He Wang. Bodex: Scalable and efficient robotic dexterous grasp synthesis using bilevel op- timization.arXiv preprint arXiv:2412.16490, 2024. 2, 3, 4

    arXiv 2024

  7. [7]

    Springgrasp: An optimization pipeline for robust and compliant dexterous pre-grasp synthesis.arXiv preprint arXiv:2404.13532, 2024

    Sirui Chen, Jeannette Bohg, and C Karen Liu. Springgrasp: An optimization pipeline for robust and compliant dexterous pre-grasp synthesis.arXiv preprint arXiv:2404.13532, 2024. 2

    arXiv 2024

  8. [8]

    Learn- ing robust real-world dexterous grasping policies via im- plicit shape augmentation.arXiv preprint arXiv:2210.13638,

    Zoey Qiuyu Chen, Karl Van Wyk, Yu-Wei Chao, Wei Yang, Arsalan Mousavian, Abhishek Gupta, and Dieter Fox. Learn- ing robust real-world dexterous grasping policies via im- plicit shape augmentation.arXiv preprint arXiv:2210.13638,

    arXiv

Show all 73 references
  1. [9]

    Syn- thesis and optimization of force closure grasps via sequential semidefinite programming.Robotics Research: Volume 1, pages 285–305, 2018

    Hongkai Dai, Anirudha Majumdar, and Russ Tedrake. Syn- thesis and optimization of force closure grasps via sequential semidefinite programming.Robotics Research: Volume 1, pages 285–305, 2018. 13

    2018

  2. [10]

    Dawson-Haggerty et al. trimesh. 14

  3. [11]

    Objaverse: A universe of annotated 3d objects

    Matt Deitke, Dustin Schwenk, Jordi Salvador, Luca Weihs, Oscar Michel, Eli VanderBilt, Ludwig Schmidt, Kiana Ehsani, Aniruddha Kembhavi, and Ali Farhadi. Objaverse: A universe of annotated 3d objects. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Reco...

    2023

  4. [12]

    Open6dor: Benchmarking open-instruction 6-dof object rearrangement and a vlm-based approach

    Yufei Ding, Haoran Geng, Chaoyi Xu, Xiaomeng Fang, Ji- azhao Zhang, Songlin Wei, Qiyu Dai, Zhizheng Zhang, and He Wang. Open6dor: Benchmarking open-instruction 6-dof object rearrangement and a vlm-based approach. In2024 IEEE/RSJ International Conference on Intelligent Robots a...

    2024

  5. [13]

    Runpei Dong, Zekun Qi, Linfeng Zhang, Junbo Zhang, Jian- jian Sun, Zheng Ge, Li Yi, and Kaisheng Ma. Autoencoders as cross-modal teachers: Can pretrained 2d image transform- ers help 3d representation learning? InThe Eleventh Inter- national Conference on Learning Representati...

    2023

  6. [14]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition ...

    2021

  7. [15]

    Graspnet-1billion: A large-scale benchmark for general ob- ject grasping

    Hao-Shu Fang, Chenxi Wang, Minghao Gou, and Cewu Lu. Graspnet-1billion: A large-scale benchmark for general ob- ject grasping. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11444– 11453, 2020. 16

    2020

  8. [16]

    Planning optimal grasps

    Carlo Ferrari, John F Canny, et al. Planning optimal grasps. InICRA, page 6, 1992. 13

    1992

  9. [17]

    Measurement of areas on a sphere using fibonacci and latitude–longitude lattices.Mathematical geo- sciences, 42:49–64, 2010

    ´Alvaro Gonz´alez. Measurement of areas on a sphere using fibonacci and latitude–longitude lattices.Mathematical geo- sciences, 42:49–64, 2010. 16

    2010

  10. [18]

    Will Grathwohl, Ricky T. Q. Chen, Jesse Bettencourt, Ilya Sutskever, and David Duvenaud. Ffjord: Free-form con- tinuous dynamics for scalable reversible generative models,

  11. [19]

    Dexfuncgrasp: A robotic dexterous functional grasp dataset constructed from a cost-effective real-simulation annotation system

    Jinglue Hang, Xiangbo Lin, Tianqiang Zhu, Xuanheng Li, Rina Wu, Xiaohong Ma, and Yi Sun. Dexfuncgrasp: A robotic dexterous functional grasp dataset constructed from a cost-effective real-simulation annotation system. InPro- ceedings of the AAAI Conference on Artificial Intelli...

    2024

  12. [20]

    Tracking ob- jects with 3d representation from videos.arXiv preprint arXiv:2306.05416, 2023

    Jiawei He, Lue Fan, Yuqi Wang, Yuntao Chen, Zehao Huang, Naiyan Wang, and Zhaoxiang Zhang. Tracking ob- jects with 3d representation from videos.arXiv preprint arXiv:2306.05416, 2023. 16

    2023 arXiv

  13. [21]

    Man- ifoldplus: A robust and scalable watertight manifold sur- face generation method for triangle soups.arXiv preprint arXiv:2005.11621, 2020

    Jingwei Huang, Yichao Zhou, and Leonidas Guibas. Man- ifoldplus: A robust and scalable watertight manifold sur- face generation method for triangle soups.arXiv preprint arXiv:2005.11621, 2020. 3, 14

    2005 arXiv

  14. [22]

    Fungrasp: Functional grasping for diverse dexter- ous hands.arXiv preprint arXiv:2411.16755, 2024

    Linyi Huang, Hui Zhang, Zijian Wu, Sammy Christen, and Jie Song. Fungrasp: Functional grasping for diverse dexter- ous hands.arXiv preprint arXiv:2411.16755, 2024. 2, 3

    2024 arXiv

  15. [23]

    Omnispatial: Towards comprehensive spatial reasoning benchmark for vi- sion language models.arXiv preprint arXiv:2506.03135,

    Mengdi Jia, Zekun Qi, Shaochen Zhang, Wenyao Zhang, Xinqiang Yu, Jiawei He, He Wang, and Li Yi. Omnispatial: Towards comprehensive spatial reasoning benchmark for vi- sion language models.arXiv preprint arXiv:2506.03135,

  16. [24]

    Hand-object contact consistency reasoning for human grasps generation

    Hanwen Jiang, Shaowei Liu, Jiashun Wang, and Xiaolong Wang. Hand-object contact consistency reasoning for human grasps generation. InProceedings of the IEEE/CVF interna- tional conference on computer vision, pages 11107–11116,

  17. [25]

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

    Moo Jin Kim, Karl Pertsch, Siddharth Karamcheti, Ted Xiao, Ashwin Balakrishna, Suraj Nair, Rafael Rafailov, Ethan Foster, Grace Lam, Pannag Sanketi, et al. Openvla: An open-source vision-language-action model.arXiv preprint arXiv:2406.09246, 2024. 2, 3

    2024 arXiv

  18. [26]

    Frogger: Fast robust grasp generation via the min-weight metric

    Albert H Li, Preston Culbertson, Joel W Burdick, and Aaron D Ames. Frogger: Fast robust grasp generation via the min-weight metric. In2023 IEEE/RSJ International Confer- ence on Intelligent Robots and Systems (IROS), pages 6809–

  19. [27]

    Multi-graspllm: A multimodal llm for multi-hand semantic guided grasp generation.arXiv preprint arXiv:2412.08468, 2024

    Haosheng Li, Weixin Mao, Weipeng Deng, Chenyu Meng, Haoqiang Fan, Tiancai Wang, Ping Tan, Hongan Wang, and Xiaoming Deng. Multi-graspllm: A multimodal llm for multi-hand semantic guided grasp generation.arXiv preprint arXiv:2412.08468, 2024. 3, 18

    2024 arXiv

  20. [28]

    Semgrasp: Semantic grasp generation via language aligned discretization

    Kailin Li, Jingbo Wang, Lixin Yang, Cewu Lu, and Bo Dai. Semgrasp: Semantic grasp generation via language aligned discretization. InEuropean Conference on Computer Vision, pages 109–127. Springer, 2025. 3, 18

    2025

  21. [29]

    Incremental potential con- tact: intersection-and inversion-free, large-deformation dy- namics.ACM Trans

    Minchen Li, Zachary Ferguson, Teseo Schneider, Timo- thy R Langlois, Denis Zorin, Daniele Panozzo, Chenfanfu Jiang, and Danny M Kaufman. Incremental potential con- tact: intersection-and inversion-free, large-deformation dy- namics.ACM Trans. Graph., 39(4):49, 2020. 4

    2020

  22. [30]

    Moka: Open-vocabulary robotic manipulation through mark-based visual prompting

    Fangchen Liu, Kuan Fang, Pieter Abbeel, and Sergey Levine. Moka: Open-vocabulary robotic manipulation through mark-based visual prompting. InFirst Workshop on Vision-Language Models for Navigation and Manipulation at ICRA 2024, 2024. 3

    2024

  23. [31]

    Openshape: Scaling up 3d shape representation towards open-world understanding.Advances in neural information processing systems, 36:44860–44879, 2023

    Minghua Liu, Ruoxi Shi, Kaiming Kuang, Yinhao Zhu, Xu- anlin Li, Shizhong Han, Hong Cai, Fatih Porikli, and Hao Su. Openshape: Scaling up 3d shape representation towards open-world understanding.Advances in neural information processing systems, 36:44860–44879, 2023. 20, 21

    2023

  24. [32]

    Partslip: Low-shot part segmentation for 3d point clouds via pretrained image- language models

    Minghua Liu, Yinhao Zhu, Hong Cai, Shizhong Han, Zhan Ling, Fatih Porikli, and Hao Su. Partslip: Low-shot part segmentation for 3d point clouds via pretrained image- language models. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 2...

    2023

  25. [33]

    Tengyu Liu, Zeyu Liu, Ziyuan Jiao, Yixin Zhu, and Song- Chun Zhu. Synthesizing diverse and physically stable grasps with arbitrary hand structures using differentiable force clo- sure estimator.IEEE Robotics and Automation Letters, 7(1): 470–477, 2021. 2, 4, 13

    2021

  26. [34]

    Dextrack: Towards generalizable neural tracking control for dexterous manipulation from human references

    Xueyi Liu, Jianibieke Adalibieke, Qianwei Han, Yuzhe Qin, and Li Yi. Dextrack: Towards generalizable neural tracking control for dexterous manipulation from human references. arXiv preprint arXiv:2502.09614, 2025. 2

    2025 arXiv

  27. [35]

    Cross-shape atten- tion for part segmentation of 3d point clouds

    Marios Loizou, Siddhant Garg, Dmitry Petrov, Melinos Averkiou, and Evangelos Kalogerakis. Cross-shape atten- tion for part segmentation of 3d point clouds. InComputer Graphics Forum, page e14909. Wiley Online Library, 2023. 2

    2023

  28. [36]

    Find any part in 3d.arXiv preprint arXiv:2411.13550, 2024

    Ziqi Ma, Yisong Yue, and Georgia Gkioxari. Find any part in 3d.arXiv preprint arXiv:2411.13550, 2024. 2

    2024 arXiv

  29. [37]

    Dex-net 2.0: Deep learning to plan robust grasps with synthetic point clouds and analytic grasp metrics, 2017

    Jeffrey Mahler, Jacky Liang, Sherdil Niyaz, Michael Laskey, Richard Doan, Xinyu Liu, Juan Aparicio Ojea, and Ken Goldberg. Dex-net 2.0: Deep learning to plan robust grasps with synthetic point clouds and analytic grasp metrics, 2017. 13

    2017

  30. [38]

    Isaac gym: High performance GPU based physics simulation for robot learning

    Viktor Makoviychuk, Lukasz Wawrzyniak, Yunrong Guo, Michelle Lu, Kier Storey, Miles Macklin, David Hoeller, Nikita Rudin, Arthur Allshire, Ankur Handa, and Gavriel State. Isaac gym: High performance GPU based physics simulation for robot learning. InNeurIPS, 2021. 2, 3, 5

    2021

  31. [39]

    Introducing gpt-4o and more tools to chatgpt free users

    OpenAI. Introducing gpt-4o and more tools to chatgpt free users. 2024. 2, 4

    2024

  32. [40]

    Contrast with reconstruct: Contrastive 3d representation learning guided by generative pretraining

    Zekun Qi, Runpei Dong, Guofan Fan, Zheng Ge, Xiangyu Zhang, Kaisheng Ma, and Li Yi. Contrast with reconstruct: Contrastive 3d representation learning guided by generative pretraining. InInternational Conference on Machine Learn- ing, pages 28223–28243. PMLR, 2023. 6

    2023

  33. [41]

    Vpp: Efficient conditional 3d generation via voxel-point pro- gressive representation.Advances in Neural Information Processing Systems, 36:26744–26763, 2023

    Zekun Qi, Muzhou Yu, Runpei Dong, and Kaisheng Ma. Vpp: Efficient conditional 3d generation via voxel-point pro- gressive representation.Advances in Neural Information Processing Systems, 36:26744–26763, 2023

    2023

  34. [42]

    Shapellm: Universal 3d object understanding for embodied interaction

    Zekun Qi, Runpei Dong, Shaochen Zhang, Haoran Geng, Chunrui Han, Zheng Ge, Li Yi, and Kaisheng Ma. Shapellm: Universal 3d object understanding for embodied interaction. InEuropean Conference on Computer Vision, pages 214–

  35. [43]

    So- far: Language-grounded orientation bridges spatial reason- ing and object manipulation.CoRR, abs/2502.13143, 2025

    Zekun Qi, Wenyao Zhang, Yufei Ding, Runpei Dong, Xin- qiang Yu, Jingwen Li, Lingyun Xu, Baoyu Li, Xialin He, Guofan Fan, Jiazhao Zhang, Jiawei He, Jiayuan Gu, Xin Jin, Kaisheng Ma, Zhizheng Zhang, He Wang, and Li Yi. So- far: Language-grounded orientation bridges spatial reaso...

    2025

  36. [44]

    Anyteleop: A general vision-based dexterous robot arm-hand teleoperation system.arXiv preprint arXiv:2307.04577, 2023

    Yuzhe Qin, Wei Yang, Binghao Huang, Karl Van Wyk, Hao Su, Xiaolong Wang, Yu-Wei Chao, and Dieter Fox. Anyteleop: A general vision-based dexterous robot arm-hand teleoperation system.arXiv preprint arXiv:2307.04577, 2023. 2

    2023 arXiv

  37. [45]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. InInternational conference on machine learning, p...

    2021

  38. [46]

    Curobo: Parallelized collision-free robot mo- tion generation

    Balakumar Sundaralingam, Siva Kumar Sastry Hari, Adam Fishman, Caelan Garrett, Karl Van Wyk, Valts Blukis, Alexander Millane, Helen Oleynikova, Ankur Handa, Fabio Ramos, et al. Curobo: Parallelized collision-free robot mo- tion generation. In2023 IEEE International Conference ...

  39. [47]

    Segment any mesh: Zero-shot mesh part segmentation via lifting segment anything 2 to 3d.arXiv preprint arXiv:2408.13679, 2024

    George Tang, William Zhao, Logan Ford, David Benhaim, and Paul Zhang. Segment any mesh: Zero-shot mesh part segmentation via lifting segment anything 2 to 3d.arXiv preprint arXiv:2408.13679, 2024. 2, 3, 4, 14

    2024 arXiv

  40. [48]

    Gemini: a family of highly capable multimodal models.arXiv preprint arXiv:2312.11805, 2023

    Gemini Team, Rohan Anil, Sebastian Borgeaud, Jean- Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, Katie Millican, et al. Gemini: a family of highly capable multimodal models.arXiv preprint arXiv:2312.11805, 2023. 2

    2023 arXiv

  41. [49]

    Octo: An open-source generalist robot policy.arXiv preprint arXiv:2405.12213, 2024

    Octo Model Team, Dibya Ghosh, Homer Walke, Karl Pertsch, Kevin Black, Oier Mees, Sudeep Dasari, Joey Hejna, Tobias Kreiman, Charles Xu, et al. Octo: An open-source generalist robot policy.arXiv preprint arXiv:2405.12213, 2024. 2, 3

    2024 arXiv

  42. [50]

    Easy and fast evaluation of grasp stability by using ellipsoidal approx- imation of friction cone

    Tokuo Tsuji, Kensuke Harada, and Kenji Kaneko. Easy and fast evaluation of grasp stability by using ellipsoidal approx- imation of friction cone. In2009 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 1830– 1837, 2009. 13

    2009

  43. [51]

    Grasp’d: Differentiable contact-rich grasp syn- thesis for multi-fingered hands

    Dylan Turpin, Liquan Wang, Eric Heiden, Yun-Chun Chen, Miles Macklin, Stavros Tsogkas, Sven Dickinson, and Ani- mesh Garg. Grasp’d: Differentiable contact-rich grasp syn- thesis for multi-fingered hands. InEuropean Conference on Computer Vision, pages 201–221. Springer, 2022. 2

    2022

  44. [52]

    Fast-grasp’d: Dexterous multi- finger grasp generation through differentiable simulation

    Dylan Turpin, Tao Zhong, Shutong Zhang, Guanglei Zhu, Eric Heiden, Miles Macklin, Stavros Tsogkas, Sven Dick- inson, and Animesh Garg. Fast-grasp’d: Dexterous multi- finger grasp generation through differentiable simulation. In 2023 IEEE International Conference on Robotics an...

    2023

  45. [53]

    Unidexgrasp++: Im- proving dexterous grasping policy learning via geometry- aware curriculum and iterative generalist-specialist learning

    Weikang Wan, Haoran Geng, Yun Liu, Zikang Shan, Yaodong Yang, Li Yi, and He Wang. Unidexgrasp++: Im- proving dexterous grasping policy learning via geometry- aware curriculum and iterative generalist-specialist learning. InProceedings of the IEEE/CVF International Conference o...

    2023

  46. [54]

    Vlm see, robot do: Human demo video to robot action plan via vision language model.arXiv preprint arXiv:2410.08792, 2024

    Beichen Wang, Juexiao Zhang, Shuwen Dong, Irving Fang, and Chen Feng. Vlm see, robot do: Human demo video to robot action plan via vision language model.arXiv preprint arXiv:2410.08792, 2024. 3

    2024

  47. [55]

    Dexcap: Scalable and portable mocap data collection system for dexterous manipulation

    Chen Wang, Haochen Shi, Weizhuo Wang, Ruohan Zhang, Li Fei-Fei, and C Karen Liu. Dexcap: Scalable and portable mocap data collection system for dexterous manipulation. arXiv preprint arXiv:2403.07788, 2024. 2

    2024 arXiv

  48. [56]

    Dexgraspnet: A large-scale robotic dexterous grasp dataset for general ob- jects based on simulation

    Ruicheng Wang, Jialiang Zhang, Jiayi Chen, Yinzhen Xu, Puhao Li, Tengyu Liu, and He Wang. Dexgraspnet: A large-scale robotic dexterous grasp dataset for general ob- jects based on simulation. In2023 IEEE International Con- ference on Robotics and Automation (ICRA), pages 11359...

    2023

  49. [57]

    Approx- imate convex decomposition for 3d meshes with collision- aware concavity and tree search.ACM Transactions on Graphics (TOG), 41(4):1–18, 2022

    Xinyue Wei, Minghua Liu, Zhan Ling, and Hao Su. Approx- imate convex decomposition for 3d meshes with collision- aware concavity and tree search.ACM Transactions on Graphics (TOG), 41(4):1–18, 2022. 3, 14

    2022

  50. [58]

    Grasp as you say: Language-guided dexterous grasp genera- tion.arXiv preprint arXiv:2405.19291, 2024

    Yi-Lin Wei, Jian-Jian Jiang, Chengyi Xing, Xian-Tuo Tan, Xiao-Ming Wu, Hao Li, Mark Cutkosky, and Wei-Shi Zheng. Grasp as you say: Language-guided dexterous grasp genera- tion.arXiv preprint arXiv:2405.19291, 2024. 2, 3, 18

    2024 arXiv

  51. [59]

    Cross- category functional grasp transfer.IEEE Robotics and Au- tomation Letters, 2024

    Rina Wu, Tianqiang Zhu, Xiangbo Lin, and Yi Sun. Cross- category functional grasp transfer.IEEE Robotics and Au- tomation Letters, 2024. 2, 3

    2024

  52. [60]

    Florence-2: Advancing a unified representation for a variety of vision tasks

    Bin Xiao, Haiping Wu, Weijian Xu, Xiyang Dai, Houdong Hu, Yumao Lu, Michael Zeng, Ce Liu, and Lu Yuan. Florence-2: Advancing a unified representation for a variety of vision tasks. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4818–...

    2024

  53. [61]

    Dexterous grasp transformer, 2024

    Guo-Hao Xu, Yi-Lin Wei, Dian Zheng, Xiao-Ming Wu, and Wei-Shi Zheng. Dexterous grasp transformer, 2024. 2

    2024

  54. [62]

    Unidexgrasp: Universal robotic dexterous grasping via learning diverse proposal generation and goal-conditioned policy

    Yinzhen Xu, Weikang Wan, Jialiang Zhang, Haoran Liu, Zikang Shan, Hao Shen, Ruicheng Wang, Haoran Geng, Yijia Weng, Jiayi Chen, et al. Unidexgrasp: Universal robotic dexterous grasping via learning diverse proposal generation and goal-conditioned policy. InProceedings of the I...

    2023

  55. [63]

    Set-of-mark prompting unleashes extraordinary visual grounding in gpt-4v.arXiv preprint arXiv:2310.11441, 2023

    Jianwei Yang, Hao Zhang, Feng Li, Xueyan Zou, Chunyuan Li, and Jianfeng Gao. Set-of-mark prompting unleashes extraordinary visual grounding in gpt-4v.arXiv preprint arXiv:2310.11441, 2023. 3, 20

    2023 arXiv

  56. [64]

    Oakink: A large-scale knowledge repos- itory for understanding hand-object interaction

    Lixin Yang, Kailin Li, Xinyu Zhan, Fei Wu, Anran Xu, Liu Liu, and Cewu Lu. Oakink: A large-scale knowledge repos- itory for understanding hand-object interaction. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 20953–20962, 2022. 2, 3

    2022

  57. [65]

    Sampart3d: Segment any part in 3d objects.arXiv preprint arXiv:2411.07184, 2024

    Yunhan Yang, Yukun Huang, Yuan-Chen Guo, Liangjun Lu, Xiaoyang Wu, Edmund Y Lam, Yan-Pei Cao, and Xihui Liu. Sampart3d: Segment any part in 3d objects.arXiv preprint arXiv:2411.07184, 2024. 2, 7

    2024 arXiv

  58. [66]

    Graspxl: Generating grasping motions for di- verse objects at scale

    Hui Zhang, Sammy Christen, Zicong Fan, Otmar Hilliges, and Jie Song. Graspxl: Generating grasping motions for di- verse objects at scale. InEuropean Conference on Computer Vision, pages 386–403. Springer, 2024. 3

    2024

  59. [67]

    Dexgrasp- net 2.0: Learning generative dexterous grasping in large- scale synthetic cluttered scenes

    Jialiang Zhang, Haoran Liu, Danshi Li, XinQiang Yu, Hao- ran Geng, Yufei Ding, Jiayi Chen, and He Wang. Dexgrasp- net 2.0: Learning generative dexterous grasping in large- scale synthetic cluttered scenes. In8th Annual Conference on Robot Learning, 2024. 2, 3, 4, 6, 9, 13, 18

    2024

  60. [68]

    Dexgrasp-diffusion: Diffusion- based unified functional grasp synthesis method for multi- dexterous robotic hands.arXiv preprint arXiv:2407.09899,

    Zhengshen Zhang, Lei Zhou, Chenchen Liu, Zhiyang Liu, Chengran Yuan, Sheng Guo, Ruiteng Zhao, Marcelo H Ang Jr, and Francis EH Tay. Dexgrasp-diffusion: Diffusion- based unified functional grasp synthesis method for multi- dexterous robotic hands.arXiv preprint arXiv:2407.09899,

    arXiv

  61. [69]

    Point transformer

    Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip HS Torr, and Vladlen Koltun. Point transformer. InProceedings of the IEEE/CVF international conference on computer vision, pages 16259–16268, 2021. 2

    2021

  62. [70]

    Transfusion: Pre- dict the next token and diffuse images with one multi- modal model

    Chunting Zhou, LILI YU, Arun Babu, Kushal Tirumala, Michihiro Yasunaga, Leonid Shamis, Jacob Kahn, Xuezhe Ma, Luke Zettlemoyer, and Omer Levy. Transfusion: Pre- dict the next token and diffuse images with one multi- modal model. InThe Thirteenth International Conference on Lea...

    2025

  63. [71]

    Uni3d: Exploring uni- fied 3d representation at scale

    Junsheng Zhou, Jinsheng Wang, Baorui Ma, Yu-Shen Liu, Tiejun Huang, and Xinlong Wang. Uni3d: Exploring uni- fied 3d representation at scale. InThe Twelfth International Conference on Learning Representations, 2024. 5, 6

    2024

  64. [72]

    Partslip++: Enhancing low-shot 3d part segmentation via multi-view instance segmenta- tion and maximum likelihood estimation.arXiv preprint arXiv:2312.03015, 2023

    Yuchen Zhou, Jiayuan Gu, Xuanlin Li, Minghua Liu, Yun- hao Fang, and Hao Su. Partslip++: Enhancing low-shot 3d part segmentation via multi-view instance segmenta- tion and maximum likelihood estimation.arXiv preprint arXiv:2312.03015, 2023. 2 A. Implementation Details of Energ...

    2023 arXiv

  65. [73]

    embedded inside

    first proposes a differentiable force closure estimator (DFC) as an energy term applicable to gradient-based opti- mization for dexterous grasp synthesis. Given a collection ofncontact force vectors, which is approximated byncon- tact normal vectorsc∈R 3n atnpositionsx∈R 3n, D...

Pith tools

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