Pith. sign in

REVIEW 3 major objections 6 minor 21 references

BiNoMaP: Learning Category-Level Bimanual Non-Prehensile Manipulation Primitives

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A single human video, geometry-corrected and resized by one size measurement, teaches a dual-arm robot to poke, pivot, push, and wrap ungraspable objects — with no reinforcement learning.

desk verdict A genuinely substantive RL-free bimanual manipulation pipeline with real-robot results and honest limitations, but the baseline comparison is stacked and the category-level adaptation rule is missing its formula. read the letter →

arxiv 2509.21256 v3 pith:B4PCF5CL submitted 2025-09-25 cs.RO

Huayi Zhou , Kui Jia This is my paper · ORCID
classification cs.RO
keywords bimanualmanipulationnon-prehensilelearningfromdemonstrationprimitivescategory-levelgeneralizationRL-freeimitationgeometry-awaretrajectoryoptimizationcross-embodimenttransfer
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

BiNoMaP sets out to show that contact-rich bimanual skills — poking, pivoting, pushing, and wrapping objects that cannot be grasped — can be learned from a single human video demonstration per skill, without reinforcement learning, simulators, or reward engineering. The two hands' motions are retargeted onto a dual-arm robot and then refined by geometry-aware smoothing and iterative real-robot contact verification into an executable primitive. Each primitive is parameterized by one object-size measurement, so the same trajectory resizes for unseen objects of the same category, and the unchanged primitive deploys on a second robot with different kinematics. If the claims hold, the practical bottleneck of non-prehensile manipulation shifts from hours of policy training to fewer than five minutes of geometric correction per skill.

What carries the argument

The argument rides on two geometric operations. The first is the geometry-aware iterative contact adjustment: with the support arm's trajectory fixed, each iteration moves the primary arm's initial contact point toward the object to a target distance $d^{(k)} = d^{(1)}\gamma^{k-1}$ and rescales the whole primary-arm trajectory by $s^{(k)} = \|p^R_{t_s,(k)} - p^L_{t_s}\|_2/\|p^R_{t_s} - p^L_{t_s}\|_2$, keeping the relative motion pattern intact until real-robot verification succeeds. The second is the category-level parameterization: the scalar $\delta_{\text{size}}$, the difference in maximum chord length between the new and base object point clouds along the line parallel to the inter-arm direction at contact height, modulates the verified contact distance and scaling factor so the primitive becomes a one-parameter family resized for each new instance. Noise reduction is carried by the coplanarity projection, cubic B-spline smoothing, and SLERP orientation interpolation, which together preserve millimeter-level contact fidelity.

What would settle it

A decisive experiment: take two upside-down bowls of identical rim diameter and identical maximum chord length along the inter-arm direction but different depth or wall steepness, so $\delta_{\text{size}} \approx 0$ for both, and run the learned pivot-bowl primitive on each. If success rates differ markedly, the scalar parameterization does not carry category-level generalization. The paper's own report that smooth ceramic, metal, and glass bowls fail even at the instance level indicates where this boundary is likely to be found.

Watch

Extended reading notes

Core claim

The paper's central claim is that a three-stage, RL-free pipeline makes bimanual non-prehensile manipulation learnable from one demonstration per skill and generalizable across a category. Stage one extracts coarse two-hand trajectories from egocentric video by reconstructing 3D hands and retargeting their contact points and orientations to parallel-jaw grippers. Stage two projects the trajectory onto an optimal plane, smooths positions with B-splines and orientations with quaternion interpolation, then iteratively rescales the moving arm's trajectory about the fixed support arm — reducing the initial contact distance geometrically until a real-robot trial succeeds. Stage three computes the size difference $\delta_{\text{size}}$ between the new object's point cloud and the base object's along the inter-arm direction at contact height and uses it to resize the verified primitive in a single step. The paper reports an 86.7% average success rate across six tasks against a 48.3% best baseline, a 76.2% category-level success rate on unseen instances, and successful cross-embodiment deployment on a second platform with a different arm configuration.

Load-bearing premise

The load-bearing premise is that a single scalar size difference — measured along the inter-arm direction at the initial contact height — fully captures how a new object of the same category differs from the one the skill was tuned on, so curvature, handle geometry, and aspect-ratio variation can be ignored.

Editorial extensions

If this is right

  • Each new non-prehensile skill costs one human demonstration plus a few minutes of contact verification — no simulator, reward function, or policy training — which makes the bottleneck of such skills cheap to clear.
  • Generalizing to a new object of the same category reduces to measuring one chord-length difference from a point cloud, so unseen instances can be attempted without repeating the demonstration or optimization stages.
  • Because a primitive is a plain trajectory rather than a learned policy, the same skill transfers across robot embodiments with different kinematics and grippers by changing only the axis convention.
  • Atomic primitives compose with vision-language localization into longer tasks: flipping an ungraspable basket so it can be grasped, stacking overturned bowls, and righting a fallen mug before pouring.
  • The reported success rates (86.7% average versus a 48.3% best baseline) indicate that explicit geometric contact reasoning can beat both visuomotor imitation and RL on contact-rich tasks with far less data.

Reading between the lines

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

  • The one-scalar recipe implies each skill behaves as a one-parameter family; the natural stress test is a two-parameter version ($\delta_{\text{size}}$ plus an orthogonal-axis delta) for categories such as boxes whose aspect ratio varies independently, and for mugs whose handle breaks the symmetry the scalar ignores.
  • Because primitives are open-loop trajectories, re-localizing the object between skill executions — already done in the downstream demonstrations — is a cheap route to closed-loop long-horizon behavior without converting each primitive into a learned controller.
  • The comparison gap over the visuomotor baselines should be read in view of the appendix's data protocol: those baselines were trained on demonstrations produced by BiNoMaP's own successful executions, so the gap measures the full pipeline's output quality as much as any difference between the policies themselves.
  • Successful transfer of the same trajectory to a humanoid-style platform with different arms suggests these primitives encode task-relevant coordination rather than arm kinematics, pointing toward shared, embodiment-independent skill libraries for non-prehensile manipulation.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes BiNoMaP, a three-stage, RL-free framework for learning bimanual non-prehensile manipulation primitives from human video demonstrations. Stage 1 extracts coarse bimanual hand trajectories using 3D hand reconstruction and retargets them to parallel-jaw grippers. Stage 2 applies geometric post-optimization: plane fitting and smoothing of positions, quaternion SLERP between anchor frames for orientations, and an iterative geometry-aware contact adjustment that scales one arm's trajectory relative to the other until a real-robot verification succeeds. Stage 3 parameterizes the resulting primitive by a scalar size difference δ_size (Eq. 5) between the base object and a new instance to support category-level generalization, plus a planar translation for new placements. The authors evaluate on four skills (poking, pivoting, pushing, wrapping) across eight object-task pairs, compare against ACT, DP, DP3, HACMan, CORN, and DyWA, report ablations, category-level generalization results, composition with VLMs and grasping for downstream tasks, and a qualitative cross-embodiment transfer experiment.

Significance. The contribution is timely and the experimental scope is substantial: real-hardware evaluation across eight tasks and four skills, two robotic platforms, a genuine hold-out category-level test, and a candid failure analysis. The category-level evaluation is a true prediction test: after one optimization on a base object, the primitive is applied to unseen object instances without retraining. The modular design is sensible, and the ablation studies support the necessity of each component. However, the central generalization mechanism (Sec. 3.4) is under-specified, and the baseline comparison is partially confounded by the training-data generation procedure described in Appendix B. These issues are fixable but currently prevent the paper's strongest claims from being fully verified. If the authors provide the explicit δ_size mapping and either re-do or re-frame the baseline comparison, the result would be a solid systems contribution.

major comments (3)
  1. [Sec. 3.4, Eq. (5)] The category-level adaptation is never specified. Equation (5) defines δ_size as the difference of maximum chord lengths of the new and base point clouds along the line parallel to the inter-arm direction, but the paper does not state how δ_size is mapped to the verified target distance d^(k) and the scaling factor s^(k) used in Eq. (4). The sentence that δ_size is "incorporated into our contact optimization Eqn. 4 in a single, non-iterative step" does not give a functional form, so a reader cannot reproduce the adapted trajectory for a new object and cannot determine whether the 76.2% category-level success rate in Table 4 is produced by the stated mechanism. This is load-bearing because the scalar parameterization also ignores shape variations within a category (bowl curvature, mug handle asymmetry, box aspect ratio); the category-level results in Table 4 are heterogeneous (e.g., 46/70 for bowls, 43/60 for mugs), which is consistent with an under-parameterized cue. The sensitivity of the mapping matters because Appendix D identifies small inter-arm distance discrepancies as the most frequent wrapping-basket failure mode. Please provide the explicit formula, including sign and magnitude handling, or revise the mechanistic explanation and present the scaling as a heuristic with empirical support.
  2. [Appendix B] The visuomotor baselines (ACT, DP, DP3) are trained on 50 demonstrations collected from successful executions of the BiNoMaP pipeline itself. This means Table 1 compares the proposed geometric optimization with policies that only observe its output trajectories, not with visuomotor policies trained in the standard way from teleoperation or human demonstrations. The main text should disclose this dependency. As written, the statement that BiNoMaP "significantly outperforms strong visuomotor baselines" overstates the evidence, because the baseline numbers may be either inflated (they receive high-quality optimized demos) or deflated (they are confined to a single strategy). Please either train the baselines on independent demonstrations or explicitly label the comparison as "imitation from BiNoMaP demonstrations" and temper the conclusions accordingly.
  3. [Appendix E / Abstract] The abstract and conclusions claim cross-embodiment transfer, but the evaluation in Appendix E is qualitative only: two skills (pivoting and wrapping) are shown on a second platform with no success-rate counts or number of trials. A claim as strong as "supports cross-embodiment transfer" requires quantitative evidence or should be revised to state that transfer was demonstrated qualitatively. Please add success rates on the Rokae platform or qualify the claim in the abstract and conclusion.
minor comments (6)
  1. [Tables 1 and 4] Success rates are reported from 10 trials per cell without confidence intervals; with n=10, a 7/10 versus 10/10 difference is not statistically significant. Please include exact binomial confidence intervals or increase the trial counts for the headline comparisons.
  2. [Sec. 2] The claim that "all existing studies adopt a single-arm setup" is contradicted by the bimanual ball-lifting and wrapping works cited in the same section (e.g., Grotz et al. 2024; Lu et al. 2025; Liu et al. 2025a). Please qualify this as "most existing studies" or otherwise acknowledge the cited bimanual exceptions.
  3. [Sec. 3.4] The phrase "scaling factors s^(k)" should be singular, since s^(k) in Eq. (4) is a single scalar scaling factor.
  4. [Sec. 3.3] The iterative contact adjustment loop depends on detecting whether "the manipulation fails," but the failure-detection procedure is not specified. Please state whether failure is judged automatically by perception or by a human operator, since the stopping criterion affects the final primitive.
  5. [Appendix A.2] The pushing-basket success criterion "without causing significant displacement" is not quantified. Please define an objective threshold (e.g., maximum displacement of the basket or bottles) for reproducibility.
  6. [Eq. (5)] The notation in the constraint "(u−v)∥(p^L_ts−p^R_ts)" uses hats that are not defined in Sec. 3.4 and the exact geometric meaning (strict parallel, or parallel after projection onto the contact plane) should be clarified.

Circularity Check

1 steps flagged · score 4.0 of 10

Visuomotor baselines are trained on BiNoMaP's own optimized trajectories, making the Tab. 1 superiority claim partly circular; the core category-level derivation is under-specified but not circular.

  1. fitted input called prediction [Appendix B, 'Reproduction of All Six Baselines'; load-bearing for Sec. 4.2 (A1) and Tab. 1.]
    "the most critical training data were derived directly from the successful demonstrations obtained through our proposed BiNoMaP pipeline. Specifically, when BiNoMaP successfully executed a given primitive on a target object, we simultaneously recorded the visual observations and the continuous 6-DoF end-effector trajectories of both arms. To ensure that the compared methods converge normally, we collected 50 demonstrations for training on the three skills and six tasks listed in Tab. 1."

    ACT, DP and DP3 are fitted to trajectories that are outputs of the three-stage BiNoMaP pipeline, then their success rates in Tab. 1 are reported as independent baselines. The comparison therefore measures BiNoMaP against supervised imitators of its own optimized output, not against independently learned non-prehensile policies; the claim that BiNoMaP 'significantly outperforms advanced non-prehensile manipulation methods' is partly forced by the way the baseline training set was constructed. The RL-based baselines and the held-out instance/category tests are not affected, so the circularity is partial.

full rationale

No equation-level circularity was found in the main derivation chain: video -> coarse trajectory -> geometry-aware optimization -> delta_size parameterization -> deployment is not self-definitional, because the final trajectory is an explicit function of a measured object attribute and the category-level tests use objects and placements not used in optimization. The Sec. 3.4 mapping from Eq. 5 to Eq. 4 is genuinely under-specified ('modulated by delta_size' without a functional form), but that is an omitted derivation and reproducibility gap, not a circular identity. The authors' own RSS 2025 paper appears only as related work and is not load-bearing. The one substantive circular element is the baseline construction in Appendix B: the three visuomotor baselines are trained on BiNoMaP-generated demonstrations, so their inferiority in Tab. 1 is partly expected by construction. Because the core hold-out generalization results and the RL-based baseline comparisons remain independent, the overall circularity score is moderate rather than severe.

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

The central claim (a generalizable RL-free bimanual non-prehensile primitive learner) rests on the coplanarity prerequisite, on the accuracy of the hand-to-point-cloud association, on VLM segmentation, and on the scalar δ_size parameterization for category transfer. The real-robot contact adjustment introduces several tuned constants (d(1), γ, top-n, gripper openings) that are fitted per task; they are ablated in Tab. 3 but remain hand-chosen. No new physical entities are introduced.

free parameters (5)
  • initial contact distance d(1) = 5 mm (default; grid-searched over 3-7 mm in Tab. 3)
    Starting value for the iterative contact adjustment in Sec. 3.3; chosen by hand and ablated. It sets the scale for the real-robot search.
  • decay factor γ = 0.85 (default; swept 0.75-0.95 in Tab. 3)
    Decay used to decrease d(k) on failure; chosen and ablated. It controls how quickly the search shrinks the contact distance.
  • anchor frame count top-n = 3 (default; swept 1-5 in Tab. 3)
    Number of intermediate frames used as orientation anchors for SLERP in Sec. 3.3; determined by ablation on the poking-bowl task.
  • gripper opening ratios per task = 0%, 25%, 50% depending on skill/task (Fig. 5 right)
    Hand-set gripper widths per task; these are task-specific constants chosen by the authors and not learned.
  • size-scaling gain for δ_size adaptation = 1 (implicit; no explicit equation given)
    Sec. 3.4 says δ_size 'modulates' the verified target distance and scaling factors but does not state the functional form; the ratio appears to be used with an implicit gain of 1, which is a free choice not validated against physics.
assumptions (5)
  • domain assumption Trajectory points remain coplanar within each arm for all four skills (coplanarity prerequisite)
    Stated in Sec. 3.3 before the smoothing optimization. All positions are projected onto a least-squares plane, so any skill requiring out-of-plane motion cannot be represented. This restricts the primitive class and the claim of generality.
  • domain assumption WiLoR-based 3D hand reconstruction, after 2D projection and point-cloud association, is accurate enough at the mm level for non-prehensile contact
    Sec. 3.2 relies on associating WiLoR contact points with the stereo point cloud because WiLoR outputs per-frame meshes without camera intrinsics; the paper states 3-5mm misalignment causes failure, so the whole pipeline depends on this association being accurate or correctable.
  • domain assumption VLM (Florence-2 + SAM2) segmentation correctly isolates the manipulated object's point cloud in all conditions
    Used in Sec. 3.2 for the base object mask and in Sec. 3.4 / Appendix C.3 for the new object's point cloud; wrong segmentation would corrupt δ_size and the planar displacement.
  • ad hoc to paper A single scalar δ_size (max chord along the inter-contact direction) is sufficient to adapt a primitive to a new object of the same category
    Eq. 5 reduces each new object to one number; shape variations such as bowl curvature, mug handle position, or box aspect ratio are ignored. This is a modeling assumption introduced specifically to make category-level generalization tractable.
  • standard math Standard mathematical tools: least-squares plane fitting, cubic B-spline smoothing, SLERP quaternion interpolation, and MANO hand model behave as expected
    Invoked in Secs. 3.2 and 3.3; these are established methods, so no independent verification is required.

how reviews work

0 comments
Cite this review

Pith. "Pith review of BiNoMaP: Learning Category-Level Bimanual Non-Prehensile Manipulation Primitives." pith.science (2026). https://pith.science/paper/B4PCF5CL

@misc{pith2026250921256,
  author       = {Pith},
  title        = {Pith review of: BiNoMaP: Learning Category-Level Bimanual Non-Prehensile Manipulation Primitives},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B4PCF5CL}},
  note         = {Machine review of arXiv:2509.21256}
}
read the original abstract

Non-prehensile manipulation, encompassing ungraspable actions such as pushing, poking, pivoting, and wrapping, remains underexplored due to its contact-rich and analytically intractable nature. We revisit this problem from two perspectives. First, instead of relying on single-arm setups or favorable environmental supports (e.g., walls or edges), we advocate a generalizable dual-arm configuration and establish a suite of Bimanual Non-prehensile Manipulation Primitives (BiNoMaP). Second, departing from prevailing RL-based approaches, we propose a three-stage, RL-free framework for learning structured non-prehensile skills. We begin by extracting bimanual hand motion trajectories from egocentric video demonstrations. Since these coarse trajectories suffer from perceptual noise and morphological discrepancies, we introduce a geometry-aware post-optimization algorithm to refine them into executable manipulation primitives consistent with predefined motion patterns. To enable category-level generalization, the learned primitives are further parameterized by object-relevant geometric attributes, primarily size, allowing adaptation to unseen instances with significant shape variations. Importantly, BiNoMaP supports cross-embodiment transfer: the same primitives can be deployed on two real-world dual-arm platforms with distinct kinematic configurations, without redesigning skill structures. Extensive real-robot experiments across diverse objects and spatial configurations demonstrate the effectiveness, efficiency, and strong generalization capability of our approach.

Figures

Figures reproduced from arXiv: 2509.21256 by the authors.

Figure 1
Figure 1. Bimanual Non-Prehensile Manipulation Primitives (BiNoMaP). (Left) We propose to extract coarse hand trajectories of non-prehensile skills from human video demonstrations, and then refine and optimize them to the dual-arm robot. These reproduced skills can be further parameterized from instance-level to category-level. (Right) We extensively validated BiNoMaP on four skills (e.g., poking, pivoting, pushing, and wrapp… view at source ↗
Figure 2
Figure 2. BiNoMaP. (1) The first stage leverages strong priors from hand demonstrations to obtain coarse dual-arm trajectories for non-prehensile tasks. (2) The second stage refines these trajecto￾ries to mitigate multi-source noise and improve execution stability. (3) The final stage generalizes learned skills to novel objects within the same category by parameterizing primitives. 3 METHOD Our proposed RL-free learning frame… view at source ↗
Figure 3
Figure 3. Examples of four non-prehensile skills instantiated with different tasks and diverse objects. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Examples of utilizing learned non-prehensile skills to boost complex manipulation tasks. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Left: The fixed-base dual-arm manipulator platform used in this research. Right: The grippers opening ratios for each skill and task. A.1 HARDWARE AND DUAL-ARM PLATFORM Our experimental platform is a rectangular table measuring approximately 110 cm in length and 70 cm …
Figure 6
Figure 6. Figure 6: The object assets involved in our selected four non-prehensile skills and eight bimanual [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: Visualization of extracted hand trajectories from recorded human demonstrations for eight [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]
Figure 8
Figure 8. Figure 8: Illustrations of motion patterns for four selected skills. Best to view after zooming in. [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]
Figure 9
Figure 9. Figure 9: Illustrations of the entire trajectory point optimization process, using skills [PITH_FULL_IMAGE:figures/full_fig_p019_9.png]
Figure 10
Figure 10. Figure 10: Illustrations for achieving instance-level and category-level generalization of learned [PITH_FULL_IMAGE:figures/full_fig_p020_10.png]
Figure 11
Figure 11. Figure 11: Qualitative real robot rollout samples of all four skills. Best to view after zooming in. [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]
Figure 12
Figure 12. Figure 12: From top to bottom and left to right, we have examples of failed cases in all four skills [PITH_FULL_IMAGE:figures/full_fig_p021_12.png]
Figure 13
Figure 13. Figure 13: Difficult examples that BioMaP cannot solve at present, especially when these objects [PITH_FULL_IMAGE:figures/full_fig_p022_13.png]
Figure 14
Figure 14. Figure 14: The another dual-arm manipulator platform used for the cross-embodiment evaluation. [PITH_FULL_IMAGE:figures/full_fig_p023_14.png]
Figure 15
Figure 15. Figure 15: Qualitative real robot rollout samples of two bimanual non-prehensile skills ( [PITH_FULL_IMAGE:figures/full_fig_p023_15.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 12 canonical work pages

  1. [1]

    Compared to conventional teleoperation data collection which is inconvenient to the non-prehensile tasks, these demonstrations offer substantially higher quality and smoother action sequences, since they originate from optimized human-guided trajectories refined via our three-stage framework. We used this dataset to train the visuomotor policies in strict...

    work page 2023

  2. [5]

    D-CODA: Diffusion for Coordinated Dual-Arm Data Augmentation

    I Liu, Chun Arthur, Jason Chen, Gaurav Sukhatme, and Daniel Seita. D-coda: Diffusion for coordi- nated dual-arm data augmentation.arXiv preprint arXiv:2505.04860, 2025a. Junjia Liu, Yiting Chen, Zhipeng Dong, Shixiong Wang, Sylvain Calinon, Miao Li, and Fei Chen. Robot cooking with stir-fry: Bimanual non-prehensile manipulation of semi-fluid objects.IEEE ...

    work page Pith review arXiv

  3. [6]

    Being-h0: Vision-language-action pretraining from large- scale human videos.arXiv preprint arXiv:2507.15597,

    Hao Luo, Yicheng Feng, Wanpeng Zhang, Sipeng Zheng, Ye Wang, Haoqi Yuan, Jiazheng Liu, Chaoyi Xu, Qin Jin, and Zongqing Lu. Being-h0: Vision-language-action pretraining from large- scale human videos.arXiv preprint arXiv:2507.15597,

    arXiv

  4. [7]

    Serl: A software suite for sample-efficient robotic reinforcement learning

    Jianlan Luo, Zheyuan Hu, Charles Xu, You Liang Tan, Jacob Berg, Archit Sharma, Stefan Schaal, Chelsea Finn, Abhishek Gupta, and Sergey Levine. Serl: A software suite for sample-efficient robotic reinforcement learning. In2024 IEEE International Conference on Robotics and Automa- tion (ICRA), pp. 16961–16969. IEEE, 2024a. Jianlan Luo, Charles Xu, Jeffrey W...

    arXiv

  5. [8]

    Planning of graspless ma- nipulation by multiple robot fingers

    Yusuke Maeda, Hirokazu Kijimoto, Yasumichi Aiyama, and Tamio Arai. Planning of graspless ma- nipulation by multiple robot fingers. InProceedings 2001 ICRA. IEEE International Conference on Robotics and Automation (Cat. No. 01CH37164), volume 3, pp. 2474–2479. IEEE,

    work page 2001

  6. [9]

    Georgios Papagiannis, Norman Di Palo, Pietro Vitiello, and Edward Johns

    doi: 10.15607/RSS.2024.XX.135. Georgios Papagiannis, Norman Di Palo, Pietro Vitiello, and Edward Johns. R+x: Retrieval and execution from everyday human videos. In2025 IEEE International Conference on Robotics and Automation (ICRA). IEEE,

    doi:10.15607/rss.2024.xx.135 2024

  7. [11]

    Learning pivoting manipulation with force and vision feedback using optimization-based demonstrations.arXiv preprint arXiv:2508.01082,

    Yuki Shirai, Kei Ota, Devesh K Jha, and Diego Romeres. Learning pivoting manipulation with force and vision feedback using optimization-based demonstrations.arXiv preprint arXiv:2508.01082,

    arXiv

  8. [12]

    Dexterous non-prehensile manipulation for ungraspable object via extrinsic dexterity.arXiv preprint arXiv:2503.23120,

    Yuhan Wang, Yu Li, Yaodong Yang, and Yuanpei Chen. Dexterous non-prehensile manipulation for ungraspable object via extrinsic dexterity.arXiv preprint arXiv:2503.23120,

    arXiv

Show all 21 references
  1. [13]

    In the wild ungraspable object picking with bimanual nonprehensile manipulation.arXiv preprint arXiv:2409.15465,

    Albert Wu and Dan Kruse. In the wild ungraspable object picking with bimanual nonprehensile manipulation.arXiv preprint arXiv:2409.15465,

    arXiv

  2. [14]

    Dexsingrasp: Learning a unified policy for dexterous object singulation and grasping in cluttered environments.arXiv preprint arXiv:2504.04516, 2025a

    13 A preprint conference paper under review Lixin Xu, Zixuan Liu, Zhewei Gui, Jingxiang Guo, Zeyu Jiang, Zhixuan Xu, Chongkai Gao, and Lin Shao. Dexsingrasp: Learning a unified policy for dexterous object singulation and grasping in cluttered environments.arXiv preprint arXiv:...

  3. [15]

    Dualafford: Learning collaborative visual affordance for dual-gripper manipulation

    Yan Zhao, Ruihai Wu, Zhehuan Chen, Yourong Zhang, Qingnan Fan, Kaichun Mo, and Hao Dong. Dualafford: Learning collaborative visual affordance for dual-gripper manipulation. InInterna- tional Conference on Learning Representations, 2023b. URLhttps://openreview.net/ forum?id=I_Y...

    arXiv

  4. [18]

    From the resulting 50 frames per trial, we manually annotate the start and end frames of each task

    Each trial lasts 5 seconds, captured at 10 frames per second, ensuring that both the initial and final hand poses remain within the camera’s field of view. From the resulting 50 frames per trial, we manually annotate the start and end frames of each task. Here, Fig. 7 visualiz...

    2023

  5. [20]

    Instead, we select the farthest point from the active end-effector on the object’s point cloud as the anchor point, which allows a stable update of Eqn

    For 3https://github.com/tonyzhaozh/act 4https://github.com/real-stanford/diffusion policy 5https://github.com/YanjieZe/3D-Diffusion-Policy 6https://github.com/HACMan-2023/HACMan 7https://github.com/iMSquared/corn 8https://github.com/jiangranlv/DyW A/ 18 A preprint conference p...

    2023

  6. [21]

    blue basket

    These rigid objects cannot withstand even visually perceptible deformations, making it challenging to balance contact distance and contact force for successful flipping, even if force sensing were available. We also tested BiNoMaP on these challenging objects, but all of them ...

    2025

  7. [2012]

    14 A preprint conference paper under review APPENDIX This appendix provides additional details to complement the main text. Sec. A (Specifications of Tasks and Setups) formalizes the definitions and experimental setups of non-prehensile tasks, of- fering a clearer understandin...

    2024

  8. [2017]

    Proximal policy optimization algorithms.arXiv preprint arXiv:1707.06347,

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms.arXiv preprint arXiv:1707.06347,

    arXiv

  9. [2019]

    Particleformer: A 3d point cloud world model for multi-object, multi-material robotic manipulation.arXiv preprint arXiv:2506.23126,

    Suning Huang, Qianzhong Chen, Xiaohan Zhang, Jiankai Sun, and Mac Schwager. Particleformer: A 3d point cloud world model for multi-object, multi-material robotic manipulation.arXiv preprint arXiv:2506.23126,

    arXiv

  10. [2020]

    Robust execution of contact-rich motion plans by hybrid force- velocity control

    Yifan Hou and Matthew T Mason. Robust execution of contact-rich motion plans by hybrid force- velocity control. In2019 International Conference on Robotics and Automation (ICRA), pp. 1933–1939. IEEE,

    1933

  11. [2023]

    Do you need a hand?–a bimanual robotic dressing assistance scheme.IEEE Transactions on Robotics, 40:1906–1919,

    Jihong Zhu, Michael Gienger, Giovanni Franzese, and Jens Kober. Do you need a hand?–a bimanual robotic dressing assistance scheme.IEEE Transactions on Robotics, 40:1906–1919,

    1906

  12. [2024]

    Peract2: Benchmark- ing and learning for robotic bimanual manipulation tasks

    Markus Grotz, Mohit Shridhar, Yu-Wei Chao, Tamim Asfour, and Dieter Fox. Peract2: Benchmark- ing and learning for robotic bimanual manipulation tasks. InCoRL 2024 Workshop on Whole-body Control and Bimanual Manipulation: Applications in Humanoids and Beyond,

    2024

  13. [2025]

    Hacman++: Spatially- grounded motion primitives for manipulation

    Bowen Jiang, Yilin Wu, Wenxuan Zhou, Chris Paxton, and David Held. Hacman++: Spatially- grounded motion primitives for manipulation. InProceedings of Robotics: Science and Systems, Delft, Netherlands, July 2024a. doi: 10.15607/RSS.2024.XX.129. Hao Jiang, Yuhai Wang, Hanyang Zh...

    doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.