Fast Estimation of Globally Optimal Independent Contact Regions for Robust Grasping and Manipulation

Harnoor Ahluwalia; Jonathan P. King; Michael Zhang; Nancy S. Pollard

REVIEW 3 major objections 4 minor 38 references

Fast Estimation of Globally Optimal Independent Contact Regions for Robust Grasping and Manipulation

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read An anytime divide-and-conquer algorithm built on incremental n-dimensional Delaunay triangulation computes independent contact regions with bounded suboptimality, achieving near-100x speedups and the first epsilon-optimal ICRs for grasps…

desk verdict Genuinely new anytime ICR algorithm with credible speedups, but the central epsilon-optimality proof is incomplete for the 4-7 contact cases it headlines. read the letter →

arxiv 2506.08856 v1 pith:PIS5CJ6J submitted 2025-06-10 cs.RO

Jonathan P. King , Harnoor Ahluwalia , Michael Zhang , Nancy S. Pollard This is my paper

classification cs.RO

keywords independentcontactregionsgraspplanningDelaunaytriangulationanytimealgorithmepsilon-optimalforceclosurequalitymetricroboticmanipulation

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 sets out to show that globally optimal independent contact regions (ICRs) — surface regions such that one contact inside each region always gives a valid grasp — can be computed quickly enough for real-time planning, even for grasps with many contacts. The proposed algorithm searches the grasp configuration space by maintaining an incremental Delaunay triangulation of invalid grasps and repeatedly growing the largest axis-aligned box around the largest empty circumsphere. The paper claims this returns an ICR whose radius is within $1/\sqrt d$ of the optimal value, where $d$ is the number of contacts, in times that are often two orders of magnitude faster than brute force. It reports the first epsilon-optimal ICRs for grasps with four or more contacts, and shows that a simple policy guided by the resulting regions tolerates object size, position, and shape errors. If these claims hold, ICRs become a practical source of guidance for grasp planning, learning, and policy transfer rather than an expensive precomputation.

What carries the argument

The central object is the independent contact region, an axis-aligned box in the order simplex $\Delta_d$ of $d$ ordered contacts on a discretized object boundary, containing only grasps classified as valid. The carrying mechanism is an incremental Delaunay triangulation of the invalid grasp points: by the Delaunay property each simplex has an empty circumsphere, so the largest empty circumsphere (clipped to the order simplex) points to the most promising empty region. The algorithm grows the largest axis-aligned hypercube centered at that circumsphere, inserts the first invalid grasp it encounters, and repeats. The optimality bound comes from the fixed ratio between the inscribed and circumscribed half-edge lengths of a hypercube, $r_- = r_+ / \sqrt d$, which gives the epsilon bound $\epsilon = \sqrt d - 1$.

What would settle it

Take a two-contact grasp configuration whose invalid-grasp set includes four points in the order simplex arranged so the largest Delaunay circumsphere encloses an invalid point lying outside the corresponding empty axis-aligned square, the standard counterexample to the empty-box-implies-empty-circumsphere lemma. If running the algorithm on this configuration returns an ICR whose radius exceeds the true largest empty square by more than the claimed $1/\sqrt d$ factor, the Proposition in Section III-C is false; if it does not, the suspected gap in the proof is not fatal for this case.

Watch

Extended reading notes

Core claim

The central discovery is algorithmic: the largest independent contact region can be found by treating invalid grasps as points in an order simplex, triangulating those points with an incremental Delaunay triangulation, and using each simplex's largest empty circumsphere as a candidate location to grow an axis-aligned hypercube of valid grasps. Inserting any invalid grasp found during growth updates the triangulation, and the process repeats until the best remaining candidate cannot improve the current solution. The paper's Proposition states that the algorithm terminates either with no grasp larger than $r_{\min}$ or with an ICR whose radius is within $1/\sqrt d$ of optimal, an epsilon-optimal solution with $\epsilon = \sqrt d - 1$. On top of this guarantee, the reported experiments show roughly 100x average speedups over a brute-force L-infinity distance transform, and the first epsilon-optimal ICRs for general grasps with four to seven contacts.

Load-bearing premise

The proof assumes that if a square (or hypercube) region contains no bad grasps, then the circle (or sphere) drawn through its corners also contains no bad grasps; this is generally false for arbitrary arrangements of bad grasps, so the claimed optimality bound rests on an unproven bridge.

Editorial extensions

If this is right

Epsilon-optimal ICRs become computable for grasps with four to seven contacts, a regime where brute-force methods ran out of memory in the reported experiments.
Because the algorithm is anytime, a robot can stop after a time budget and still receive the current best ICR together with a bound on how far it is from optimal.
The roughly 100x speedup over brute force brings ICR computation into the range of real-time planning for the tested planar cases.
A control policy that simply moves toward the computed ICRs tolerated ±20% object size changes, 1.8 cm position shifts, and geometric differences in the reported hardware experiments.
Using the same quality metric to classify valid versus invalid grasps turns a point-based metric into a volume-based robustness measure, which the paper observes picks contacts that are qualitatively safer under uncertainty.

Reading between the lines

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

Because the algorithm only needs a validity oracle, it could in principle run with learned or simulation-based grasp classifiers, not just analytic force-closure metrics; the paper does not test this combination.
The appendix's proposed 3D extension replaces hypercube growth with circles on flattened meshes; implementing it and checking whether the $\epsilon$ bound survives non-Euclidean region geometry is a direct testable next step.
The same divide-and-conquer search over incrementally discovered invalid points could be adapted to other high-dimensional empty-region problems, such as collision-free configuration-space boxes for motion planning, if the proof gap around empty circumspheres is repaired.
Because the guaranteed bound weakens as $\sqrt d$ grows with contact count, practical use at large $d$ will likely rely more on the anytime behavior and early-exit heuristic than on the formal guarantee.

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.

Desk Editor's Note

read the letter

Punchline: the algorithm is genuinely new and the empirical work is credible, but the paper's central epsilon-optimality proof doesn't cover the cases it headlines. The 1/sqrt(d) guarantee is not established for 4-7 contacts because Section III-D only proves boundary-walking candidate convergence for two-contact grasps.

What's actually good: adapting incremental Delaunay triangulation to ICR search with an anytime scheme is a new idea. Prior ICR algorithms were stuck at 2-4 contacts or gave local regions. The speedups over brute force (roughly 100x at higher contact counts) are plausible, and the anytime plots are informative. The comparison against Phoka et al. reproduces similar contact regions, and the use of brute-force distance transforms as ground truth for 2-4 contacts is the right sanity check. The policy robustness experiments are qualitative but supportive.

Soft spots: The proof of the Proposition in Section III-C relies on r+ being the largest empty hypersphere in the order simplex. The stress-test note is right: for d>=3 nothing in the paper proves that the boundary-walking candidate in Section III-D is a global upper bound. The paper itself admits monotonic convergence only for two contacts, so the bounded-suboptimality claim overreaches for the 4-7 contact results. The reader's specific objection about the 'empty box's circumsphere' lemma is actually a misreading: if 'inscribed circumsphere' means the insphere of the box, that is contained and empty, so the lemma is salvageable. The real hole is the candidate upper bound, not that lemma. Also, Table I contradicts the claim that 'all of our times improved' (e.g., figure a at 10deg has t_o=1.1s vs t_p=1.02s). The 5-7 contact results have no brute-force ground truth (brute force OOM), so the reported epsilon values there are unverified. Code is not yet released, though the abstract promises it.

Summary: This is for researchers who care about grasp quality metrics, anytime planning, or robust manipulation. The algorithmic idea is worth building on, and the empirical evaluation is honest enough to engage with. But the central theoretical claim should not be taken at face value. A referee should ask for a correct proof of the candidate bound for d>=3 (possibly via Toussaint's location theorem plus the L-infinity/Euclidean ratio) and a corrected table statement. I'd send it to peer review with major revision expected, not desk reject.

Referee Report

3 major / 4 minor

Summary. The paper proposes an anytime, divide-and-conquer algorithm for computing independent contact regions (ICRs) in grasp configuration space. It models ICRs as axis-aligned boxes in an order simplex, treats invalid grasps as points, and uses an incrementally updated Delaunay triangulation to identify the largest empty circumsphere, grow a hypercube from its center, insert any newly discovered invalid grasp, and repeat. The authors claim a bounded-suboptimality guarantee (within 1/sqrt(d) of optimal in hypercube radius), report speedups of roughly 100x over a brute-force L-infinity distance transform, compare favorably with the two-contact algorithm of Phoka et al. [24], and demonstrate robustness of an ICR-guided policy to size, position, and geometry perturbations. The central contribution is the first claimed epsilon-optimal ICR computation for grasps with four or more contacts.

Significance. If the theoretical guarantee is established, this is a substantial contribution: it would provide the first epsilon-optimal ICR algorithm for high-contact grasps, with an anytime formulation and empirical speedups that make the computation practical for real-time planning. The experimental work is a clear strength: the algorithm is benchmarked against external references (brute-force L-infinity distance transforms for 2-4 contacts and the published examples of Phoka et al. [24]), and the reported speedups and anytime behavior are credible and well documented. The robustness experiments and the comparison with the Ferrari-Canny point metric are also valuable. However, the paper's central epsilon-optimality proof is incomplete for the headline d>=3 cases, so the theoretical claim currently outruns what is proven.

major comments (3)

[III-C and III-D] The Proposition in Section III-C relies on r+ being the radius of the largest empty hypersphere among all centers in the order simplex. This is not established for d>=3. Delaunay circumcenters lie in the interior of the triangulation, while for a bounded domain with a location constraint the largest empty circle can be centered on the boundary. Section III-D explicitly states that the boundary-walking procedure 'guarantees monotonic convergence of candidate scores for two contact grasps' and that for grasps with greater than two contacts it 'provides a significant practical speedup to the base algorithm.' Thus for the 4-7 contact regime -- precisely the paper's claimed first epsilon-optimal results -- the proof never shows that BestCandidate returns a global upper bound r+ on all empty spheres. Without this bridge, the lemma 'There can be no empty axis aligned hypercube with radius greater than r+' is unsupported, and the 1/sqrt(d) bound is unproven. This is the load-bearing step of the central claim and needs a rigorous proof or a revised, weaker statement.
[III-C, Lemma proof] The proof of the lemma 'There can be no empty axis aligned hypercube with radius greater than r+' uses the phrase 'inscribed circumsphere.' If this means the sphere through the hypercube's vertices (the circumscribed sphere), the statement is false: an empty axis-aligned box can have invalid grasps inside the circumsphere but outside the box, so the circumsphere need not be empty. If the intended object is the insphere (the sphere inscribed in the box, tangent to its faces), then it is contained in the box and is empty, and the argument goes through. The paper should define this notion explicitly and make the terminology consistent with Figure 3, where the same object is called the 'inscribed hypercube.' As written, a load-bearing proof step is ambiguous and, on the literal reading, incorrect.
[III-C, Case 2 and termination] The second lemma in the Proposition (that an explored simplex yields a hypercube of radius r- - epsilon_r < r_b < r+) is only sketched with 'By arguments outlined earlier in the manuscript.' The argument appears to require that when the hypercube is grown from the center of an empty circumsphere, no invalid grasp is encountered before the hypercube reaches the insphere radius r-, which is plausible because the insphere is contained in the empty circumsphere. However, the paper does not specify the exact growth procedure or prove that the first invalid grasp encountered always occurs at radius >= r-. This step should be made precise, especially because the final bound r_max >= r+/sqrt(d) depends on it.

minor comments (4)

[III-C, Proposition] The phrase 'within 1√d of optimal' should read 'within 1/sqrt(d) of optimal'; the missing slash appears in the Proposition statement and in the Early Exit paragraph of Section III-D.
[III-B, Equations (3)-(6)] The quantities r_b, r_+, r_-, and epsilon_r are used in Equations (3)-(6) with incomplete definitions; in particular, epsilon_r first appears without explanation of its role as a small positive quantity, and r_b is only later described as the half-edge length. Please define all symbols before or immediately after the display.
[III-C, Lemma proof] The notation 'b ⊂ s' and 'b̸⊂s' is inappropriate for a point and a sphere; it should be 'b ∈ s' or 'b lies inside/outside the circumsphere.' The current notation makes the proof harder to follow.
[Table II and Figure 5] The table caption should define the 'eps' column (the fractional suboptimality bound from Algorithm 1) and explain 'OOM' as 'out of memory.' In Figure 5, the left plot uses a log scale, and the dashed/solid distinction between brute force and iterative runtimes should be stated in the caption.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central algorithmic claim is benchmarked against external brute-force and published results, and the main weakness is a proof gap for d>=3 rather than a circular reduction.

full rationale

The paper's central derivation is not circular. ICRs are defined as axis-aligned boxes in an order simplex, and the algorithm iteratively maintains a Delaunay triangulation of invalid grasps while growing hypercubes around empty circumspheres. The claimed bounded-suboptimality result rests on geometric relationships such as r- = r+/sqrt(d) and on lemmas relating the largest empty hypercube to the largest empty hypersphere; these are substantive geometric claims, not tautologies. No parameter is fitted to a subset of data and then reported as a prediction: Table II compares the iterative algorithm to a brute-force L-infinity distance transform over the full grasp space, and Table I compares against the published examples of Phoka et al. [24]. The citations to Pollard's earlier ICR work ([25], [26]) appear only as related work and are not load-bearing for the algorithm's correctness or its epsilon bound. The Toussaint citation [37] is external and supplies the boundary-case guarantee for the two-contact case; for more than two contacts the paper explicitly states only that the practice provides a significant practical speedup, not a proof. That is a genuine correctness and evidence gap for the stated Proposition in the d>=3 regime, but it is not circularity: the proposition is not true by construction, no fitted value is renamed as a prediction, and no self-citation chain forces the conclusion. The empirical comparisons provide independent external anchors for the algorithm's performance claims.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central proof rests on two geometric claims: the Toussaint location theorem for largest empty circles, and an unproven, in general false relationship between empty boxes and empty circumspheres. The latter is flagged as ad hoc because it is introduced specifically to derive the epsilon bound and is not established in the text.

free parameters (2)

r_min = user-set
Termination threshold in Algorithm 1; a hand-chosen minimum acceptable ICR radius, not fitted to data.
resolution L = 32-512 samples in experiments
Discretization of the object perimeter in the experiments; a user-selected modeling parameter that affects runtime and discretization error.

assumptions (4)

standard math The largest empty Euclidean circle constrained to a convex polygon is centered at a Voronoi vertex or at a Voronoi edge intersection with the polygon boundary (Toussaint [37]).
Used in Section III-D to justify restricting candidate centers to Delaunay circumcenters and hull intersections.
ad hoc to paper An empty axis-aligned box has an empty Euclidean circumsphere, so the Delaunay largest empty circumsphere bounds the largest empty box.
Assumed in the proof of the bounding lemma in Section III-C; this is generally false (invalid points can lie in the circumsphere outside the box). The epsilon-optimality proof as written depends on it.
domain assumption Grasp validity is a binary predicate on the order simplex, and the configuration space is the order simplex of a discretized planar boundary.
Defines the problem; limiting to 2D planar objects with ordered contacts.
standard math The radius of the largest axis-aligned box centered at a point is its L-infinity distance to the nearest invalid grasp.
Geometry of L-infinity balls; used implicitly in hypercube growing.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Fast Estimation of Globally Optimal Independent Contact Regions for Robust Grasping and Manipulation." pith.science (2026). https://pith.science/paper/PIS5CJ6J

@misc{pith2026250608856,
  author       = {Pith},
  title        = {Pith review of: Fast Estimation of Globally Optimal Independent Contact Regions for Robust Grasping and Manipulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PIS5CJ6J}},
  note         = {Machine review of arXiv:2506.08856}
}

read the original abstract

This work presents a fast anytime algorithm for computing globally optimal independent contact regions (ICRs). ICRs are regions such that one contact within each region enables a valid grasp. Locations of ICRs can provide guidance for grasp and manipulation planning, learning, and policy transfer. However, ICRs for modern applications have been little explored, in part due to the expense of computing them, as they have a search space exponential in the number of contacts. We present a divide and conquer algorithm based on incremental n-dimensional Delaunay triangulation that produces results with bounded suboptimality in times sufficient for real-time planning. This paper presents the base algorithm for grasps where contacts lie within a plane. Our experiments show substantial benefits over competing grasp quality metrics and speedups of 100X and more for competing approaches to computing ICRs. We explore robustness of a policy guided by ICRs and outline a path to general 3D implementation. Code will be released on publication to facilitate further development and applications.

Figures

Figures reproduced from arXiv: 2506.08856 by the authors.

**Figure 2.** (Left) A shape defined by a curve parameterized on [0, L] with frictional contacts and contact regions. (Middle) The grasp configuration space as [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 3.** Each simplex in the Delaunay Triangulation has a candidate score. [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Illustration of iterative algorithm for computing Independent Contact Regions. (Top) A single iteration consists of identifying the best next [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

**Figure 5.** Figure 5: Left to right: (1) Contact regions Phoka vs. ours are similar; (2) our runtimes (solid) show large improvements over brute force (dashed); (3) [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

**Figure 6.** Figure 6: (Top) Optimal grasps found using the point based metric in [12]. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

38 extracted references · 37 canonical work pages

[24]

Optimal independent contact regions for two-fingered grasping of polygon,

T. Phoka, P. V ongmasa, C. Nilwatchararang, P. Pipattanasomporn, and A. Sudsang, “Optimal independent contact regions for two-fingered grasping of polygon,”Robotica, vol. 30, no. 6, pp. 879–889, 2012

work page 2012
[1]

Robotic grasping and contact: A review,

A. Bicchi and V . Kumar, “Robotic grasping and contact: A review,” in Proceedings IEEE international conference on robotics and automa- tion, vol. 1. IEEE, 2000, pp. 348–353

work page 2000
[2]

Trends and challenges in robot manipula- tion,

A. Billard and D. Kragic, “Trends and challenges in robot manipula- tion,”Science, vol. 364, no. 6446, 2019

work page 2019
[3]

Autogpt+ p: Affordance- based task planning using large language models,

T. Birr, C. Pohl, A. Younes, and T. Asfour, “Autogpt+ p: Affordance- based task planning using large language models,”Robotics: Science and Systems, 2024

work page 2024
[4]

Data-driven grasp synthesis—a survey,

J. Bohg, A. Morales, T. Asfour, and D. Kragic, “Data-driven grasp synthesis—a survey,”IEEE Transactions on robotics, vol. 30, no. 2, pp. 289–309, 2013

work page 2013
[5]

ConstructingL ∞ voronoi diagrams in 2d and 3d,

D. R. Bukenberger, K. Buchin, and M. Botsch, “ConstructingL ∞ voronoi diagrams in 2d and 3d,” inComputer Graphics Forum, vol. 41, no. 5. Wiley Online Library, 2022, pp. 135–147

work page 2022
[6]

Multi-fingered dynamic grasping for unknown objects,

Y . Burkhardt, Q. Feng, J. Feng, K. Sharma, Z. Chen, and A. Knoll, “Multi-fingered dynamic grasping for unknown objects,” in2024 IEEE-RAS 23rd International Conference on Humanoid Robots (Hu- manoids). IEEE, 2024, pp. 60–67

work page 2024
[7]

Delaunay triangulations of closed eu- clidean d-orbifolds,

M. Caroli and M. Teillaud, “Delaunay triangulations of closed eu- clidean d-orbifolds,”Discrete & Computational Geometry, vol. 55, pp. 827–853, 2016

work page 2016

Show all 38 references

[8]

Independent contact regions based on a patch contact model,

K. Charusta, R. Krug, D. Dimitrov, and B. Iliev, “Independent contact regions based on a patch contact model,” in2012 IEEE International Conference on Robotics and Automation. IEEE, 2012, pp. 4162–4169

2012
[9]

Fast and flexible determination of force- closure independent regions to grasp polygonal objects,

J. Cornella and R. Su ´arez, “Fast and flexible determination of force- closure independent regions to grasp polygonal objects,” inProceed- ings of the 2005 IEEE International Conference on Robotics and Automation. IEEE, 2005, pp. 766–771

2005
[10]

On the largest empty axis-parallel box amidst n points,

A. Dumitrescu and M. Jiang, “On the largest empty axis-parallel box amidst n points,”Algorithmica, vol. 66, no. 2, pp. 225–248, 2013

2013
[11]

Dexgangrasp: Dexterous generative adversarial grasping synthesis for task-oriented manipulation,

Q. Feng, D. S. M. Lema, M. Malmir, H. Li, J. Feng, Z. Chen, and A. Knoll, “Dexgangrasp: Dexterous generative adversarial grasping synthesis for task-oriented manipulation,” inIEEE-RAS International Conference on Humanoid Robots (Humanoids), 2024

2024
[12]

Planning optimal grasps

C. Ferrari, J. F. Canny,et al., “Planning optimal grasps.” in1992 IEEE International Conference on Robotics and Automation, vol. 3, no. 4, 1992, pp. 2290–2295

1992
[13]

Integrated grasp and motion planning using independent contact regions,

J. Fontanals, B.-A. Dang-Vu, O. Porges, J. Rosell, and M. A. Roa, “Integrated grasp and motion planning using independent contact regions,” in2014 IEEE-RAS International Conference on Humanoid Robots. IEEE, 2014, pp. 887–893

2014
[14]

Projections onto order simplexes,

S. J. Grotzinger and C. Witzgall, “Projections onto order simplexes,” Applied mathematics and Optimization, vol. 12, pp. 247–270, 1984

1984
[15]

Structural synthesis and optimisation of a robotic gripper using generative ai design,

H. Isakhani, S. Nefti-Meziani, S. Davis, A. M. Hajiyavand, and X. Xu, “Structural synthesis and optimisation of a robotic gripper using generative ai design,” inIEEE-RAS 23rd International Conference on Humanoid Robots (Humanoids), 2024, pp. 91–97

2024
[16]

In-hand rolling motion planning using independent contact region (ICR) with guaranteed grasp qu ality margin,

H. Jeong and J. Cheong, “In-hand rolling motion planning using independent contact region (ICR) with guaranteed grasp qu ality margin,” in2013 IEEE International Conference on Robotics and Automation. IEEE, 2013, pp. 3239–3244

2013
[17]

Adaptive fingers coordination for robust grasp and in-hand manipulation under disturbances and unknown dynamics,

F. Khadivar and A. Billard, “Adaptive fingers coordination for robust grasp and in-hand manipulation under disturbances and unknown dynamics,”IEEE Transactions on Robotics, vol. 39, no. 5, 2023

2023
[18]

On the efficient computation of independent contact regions for force closure grasps,

R. Krug, D. Dimitrov, K. Charusta, and B. Iliev, “On the efficient computation of independent contact regions for force closure grasps,” in2010 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, 2010, pp. 586–591

2010
[19]

Big holes in big data: A monte carlo algorithm for detecting large hyper-rectangles in high dimensional data,

J. Lemley, F. Jagodzinski, and R. Andonie, “Big holes in big data: A monte carlo algorithm for detecting large hyper-rectangles in high dimensional data,” inIEEE 40th annual computer software and applications conference (COMPSAC), vol. 1, 2016, pp. 563–571

2016
[20]

Leveraging pre- trained latent representations for few-shot imitation learning on an anthropomorphic robotic hand,

D. Liconti, Y . Toshimitsu, and R. Katzschmann, “Leveraging pre- trained latent representations for few-shot imitation learning on an anthropomorphic robotic hand,” inIEEE-RAS 23rd International Con- ference on Humanoid Robots (Humanoids), 2024, pp. 181–188

2024
[21]

The synthesis of stable force-closure grasps,

V .-D. Nguyen, “The synthesis of stable force-closure grasps,” MS Thesis, MIT, Cambridge, MA, May 1986

1986
[22]

Constructing force-closure grasps,

——, “Constructing force-closure grasps,”The International Journal of Robotics Research, vol. 7, no. 3, pp. 3–16, 1988

1988
[23]

Regrasp planning for a 5-fingered hand manipulating a polyhedron,

T. Phoka and A. Sudsang, “Regrasp planning for a 5-fingered hand manipulating a polyhedron,” inProceedings IEEE/RSJ International Conference on Intelligent Robots and Systems, vol. 4, 2003, pp. 3674– 3679

2003
[25]

Parallel methods for synthesizing whole-hand grasps from generalized prototypes,

N. S. Pollard, “Parallel methods for synthesizing whole-hand grasps from generalized prototypes,” PhD thesis, Massachusetts Institute of Technology, Cambridge, MA, 1994, department of EECS

1994
[26]

Closure and quality equivalence for efficient synthesis of grasps from examples,

——, “Closure and quality equivalence for efficient synthesis of grasps from examples,”The International Journal of Robotics Research, vol. 23, no. 6, pp. 595–613, 2004

2004
[27]

On computing three-finger force-closure grasps of polygonal objects,

J. Ponce and B. Faverjon, “On computing three-finger force-closure grasps of polygonal objects,”IEEE Transactions on robotics and automation, vol. 11, no. 6, pp. 868–881, 1995

1995
[28]

On computing two-finger force- closure grasps of curved 2d objects,

J. Ponce, D. Stam, and B. Faverjon, “On computing two-finger force- closure grasps of curved 2d objects,”The International Journal of Robotics Research, vol. 12, no. 3, pp. 263–273, 1993

1993
[29]

On computing four-finger equilibrium and force-closure grasps of polyhedral objects,

J. Ponce, S. Sullivan, A. Sudsang, J.-D. Boissonnat, and . J.-P. Merlet, “On computing four-finger equilibrium and force-closure grasps of polyhedral objects,”The International Journal of Robotics Research, vol. 16, no. 1, pp. 11–35, 1997

1997
[30]

Computation of independent contact regions for grasping 3-d objects,

M. A. Roa and R. Su ´arez, “Computation of independent contact regions for grasping 3-d objects,”IEEE Transactions on Robotics, vol. 25, no. 4, pp. 839–850, 2009

2009
[31]

Influence of contact types and uncertainties in the computation of independent contact regions,

——, “Influence of contact types and uncertainties in the computation of independent contact regions,” in2011 IEEE International Confer- ence on Robotics and Automation. IEEE, 2011, pp. 3317–3323

2011
[32]

An overview of 3d object grasp synthesis algorithms,

A. Sahbani, S. El-Khoury, and P. Bidaud, “An overview of 3d object grasp synthesis algorithms,”Robotics and Autonomous Systems, vol. 60, no. 3, pp. 326–336, 2012

2012
[33]

Apricot: Action primitives based on contact-state tran- sition for in-hand tool manipulation,

D. Saito, A. Kanehira, K. Sasabuchi, N. Wake, J. Takamatsu, H. Koike, and K. Ikeuchi, “Apricot: Action primitives based on contact-state tran- sition for in-hand tool manipulation,” inIEEE-RAS 23rd International Conference on Humanoid Robots (Humanoids), 2024, pp. 827–834

2024
[34]

Boundary first flattening,

R. Sawhney and K. Crane, “Boundary first flattening,”ACM Transac- tions on Graphics (ToG), vol. 37, no. 1, pp. 1–14, 2017

2017
[35]

Deep reinforcement learning for robotics: A survey of real- world successes,

C. Tang, B. Abbatematteo, J. Hu, R. Chandra, R. Mart ´ın-Mart´ın, and P. Stone, “Deep reinforcement learning for robotics: A survey of real- world successes,”arXiv preprint arXiv:2408.03539, 2024

2024 arXiv
[36]

Words2contact: Identifying support contacts from verbal instructions using founda- tion models,

D. Totsila, Q. Rouxel, J.-B. Mouret, and S. Ivaldi, “Words2contact: Identifying support contacts from verbal instructions using founda- tion models,” in2024 IEEE-RAS 23rd International Conference on Humanoid Robots (Humanoids). IEEE, 2024, pp. 9–16

2024
[37]

Computing largest empty circles with location con- straints,

G. T. Toussaint, “Computing largest empty circles with location con- straints,”International journal of computer & information sciences, vol. 12, pp. 347–358, 1983

1983
[38]

Multi-finger manipulation via trajectory optimization with differentiable rolling and geometric constraints,

F. Yang, T. Power, S. A. Marinovic, S. Iba, R. S. Zarrin, and D. Berenson, “Multi-finger manipulation via trajectory optimization with differentiable rolling and geometric constraints,”IEEE Robotics and Automation Letters, 2025

2025

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [24]

Optimal independent contact regions for two-fingered grasping of polygon,

T. Phoka, P. V ongmasa, C. Nilwatchararang, P. Pipattanasomporn, and A. Sudsang, “Optimal independent contact regions for two-fingered grasping of polygon,”Robotica, vol. 30, no. 6, pp. 879–889, 2012

work page 2012

[2] [1]

Robotic grasping and contact: A review,

A. Bicchi and V . Kumar, “Robotic grasping and contact: A review,” in Proceedings IEEE international conference on robotics and automa- tion, vol. 1. IEEE, 2000, pp. 348–353

work page 2000

[3] [2]

Trends and challenges in robot manipula- tion,

A. Billard and D. Kragic, “Trends and challenges in robot manipula- tion,”Science, vol. 364, no. 6446, 2019

work page 2019

[4] [3]

Autogpt+ p: Affordance- based task planning using large language models,

T. Birr, C. Pohl, A. Younes, and T. Asfour, “Autogpt+ p: Affordance- based task planning using large language models,”Robotics: Science and Systems, 2024

work page 2024

[5] [4]

Data-driven grasp synthesis—a survey,

J. Bohg, A. Morales, T. Asfour, and D. Kragic, “Data-driven grasp synthesis—a survey,”IEEE Transactions on robotics, vol. 30, no. 2, pp. 289–309, 2013

work page 2013

[6] [5]

ConstructingL ∞ voronoi diagrams in 2d and 3d,

D. R. Bukenberger, K. Buchin, and M. Botsch, “ConstructingL ∞ voronoi diagrams in 2d and 3d,” inComputer Graphics Forum, vol. 41, no. 5. Wiley Online Library, 2022, pp. 135–147

work page 2022

[7] [6]

Multi-fingered dynamic grasping for unknown objects,

Y . Burkhardt, Q. Feng, J. Feng, K. Sharma, Z. Chen, and A. Knoll, “Multi-fingered dynamic grasping for unknown objects,” in2024 IEEE-RAS 23rd International Conference on Humanoid Robots (Hu- manoids). IEEE, 2024, pp. 60–67

work page 2024

[8] [7]

Delaunay triangulations of closed eu- clidean d-orbifolds,

M. Caroli and M. Teillaud, “Delaunay triangulations of closed eu- clidean d-orbifolds,”Discrete & Computational Geometry, vol. 55, pp. 827–853, 2016

work page 2016