Pith. sign in

REVIEW 4 major objections 5 minor 36 references

KnotDLO: Toward Interpretable Knot Tying

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Training-free robot ties overhand knots at 50 percent

desk verdict Genuine integration but the headline success leans on a geometric fallback at the weakest move; worth a careful review with trimmed claims. read the letter →

arxiv 2506.22176 v1 pith:WJ5GB3EZ submitted 2025-06-27 cs.RO cs.CV

Holly Dinkel , Raghavendra Navaratna , Jingyi Xiang , Brian Coltin , Trey Smith , Timothy Bretl This is my paper · ORCID
classification cs.ROcs.CV
keywords knottyingdeformablelinearobjectstopologicalwaypointsReidemeistermovesvisualtrackingunderocclusioninterpretablerobotmanipulationtraining-freeplanning
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

KnotDLO sets out to show that robotic knot tying does not need demonstrations, training, or a learned dynamics model: a planner reads the current tracked shape of a rope, converts that shape into a small set of grasp and target waypoints, and moves the rope through the topology moves that form an overhand knot. The rope is tracked as a piecewise linear curve, and every waypoint is derived from the curve's geometry and topology, so the policy is interpretable and does not depend on rope length or the exact starting configuration. In 16 trials from previously unseen configurations, the system tied the knot 8 times, a 50% success rate, with per-move success rates of 93.7%, 86.7%, and 61.5% for the three moves. The authors state this is the first knot-tying method to use topological deformable-linear-object state tracking from a perception system as its input. If the claim stands, a useful but imperfect knot-tying capability can be assembled from perception plus geometry, with failures traceable to a specific move rather than hidden inside a learned model.

What carries the argument

The central object is the tracked piecewise linear curve $S$, a rope-shaped curve built from $M$ control points and parameterized by curvilinear length, together with a geodesic distance $\rho(s_i,s_j)$ that measures separation along the rope rather than through 3D space. This curve is the single source of truth for planning: grasp positions are curve points, grasp orientations come from the curve's unit tangent under the semi-planar constraint, and target positions are computed from geometric combinations of selected curve points. The movement primitives are the Reidemeister moves from knot theory (twist, slide-over-loop, slide-over-crossing) plus a Cross Move for open ropes, and the planner selects seven special curvilinear lengths -- the midpoint, the two symmetric quarter points, the crossing bottom and top, and the under-tip and over-tip -- to turn each primitive into concrete waypoints. The mechanism does its work by decoupling visual reasoning from control: whatever the tracking system estimates about the rope's shape and self-crossings is converted deterministically into robot motion, which is why the authors call the result interpretable.

What would settle it

Use a second overhead camera to record the true self-crossing order during the transition from the second move to the final poke-through move, and correlate each trial's final-move outcome with whether the tracked curve had the correct crossing topology; the central claim predicts that final moves succeed only in trials where the tracked topology was correct.

Watch

Extended reading notes

Core claim

The central claim is that an overhand knot can be tied by composing three movement primitives -- Reidemeister Move I, which twists the rope to add one crossing; Reidemeister Move II, which slides one strand over a loop to add two crossings; and the Cross Move, which pokes a rope tip through a loop to finish the knot -- and that the grasp and target poses for each primitive can be computed directly from the currently tracked piecewise linear curve $S(s_t;L)$ with no training and no human demonstrations. The planner selects seven curvilinear lengths on the curve, including the middle point, the two symmetric quarter points, and the crossing bottom and top, and turns them into waypoints using the local tangent direction under a semi-planar manipulation constraint. The authors report that this recipe tied an overhand knot in 8 of 16 trials from previously unseen symmetric curved starting configurations, and they identify the final Cross Move as the fragile step. They also state that KnotDLO is the first knot-tying system that takes topological DLO state tracking from a perception system as its input, which is what makes the waypoint plan interpretable rather than learned.

Load-bearing premise

The planner assumes the tracked rope shape tells the truth about which strand is on top at the crossing the final move must poke through, and the paper itself notes that this estimate often fails after the second move because depth resolution is too coarse.

Editorial extensions

If this is right

  • Because waypoints are computed from the tracked curve rather than learned, the same planner transfers to new rope lengths and new initial symmetric configurations without retraining.
  • The reported per-move success rates put the bottleneck on the final Cross Move (61.5%), so improving crossing-bottom localization and under-tip selection should raise overall success more than improving the earlier moves.
  • Decoupling visual reasoning from control means improvements in occlusion-robust rope tracking can be adopted without changing the planner.
  • The interpretable waypoint sequence can be used to collect labeled demonstration data for learning-based manipulation policies, which the authors propose as future work.

Reading between the lines

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

  • Beyond the reported trials, the total success rate 0.500 is nearly the product of the per-move rates ($0.937 \times 0.867 \times 0.615 \approx 0.500$), consistent with move failures being roughly independent; if that independence holds, raising the Cross Move alone to 90% success would lift overall success to about 73%.
  • A test the paper does not run: vary depth noise or occlusion level while using a second camera to record the true self-crossing order, separating perception errors from execution errors and directly testing whether tracking accuracy is the load-bearing factor.
  • The same geometric recipe -- choose a knot type, choose a sequence of Reidemeister moves and Cross Moves, compute all waypoints from the tracked curve -- should extend to other knots such as the figure-eight, since the primitives are not specific to the overhand knot.
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

4 major / 5 minor

Summary. The paper presents KnotDLO, a one-handed overhand knot tying system that uses TrackDLO's real-time piecewise-linear DLO tracking to compute grasp and target waypoints from the current DLO shape, without learning from demonstrations. The planner composes three primitives—Reidemeister Move I, Reidemeister Move II, and the Cross Move—based on the tracked topology and geometry. In 16 trials from previously unseen symmetric curved configurations, the system tied an overhand knot in 8 trials (50% success), with per-move success rates of 0.937, 0.867, and 0.615. The authors claim this is the first knot tying method using topological DLO state tracking from a perception system as input, and they emphasize interpretability, repeatability, and occlusion robustness.

Significance. If the reported result is robust, the contribution is notable: a training-free, interpretable alternative to learned knotting policies, with an explicit mapping from tracked DLO topology to waypoint poses and an empirical evaluation rather than a fitted prediction loop. The 50% success rate is honestly positioned against a learned baseline (66% for DDOD), and the per-move breakdown is a useful diagnostic. The significance is tempered, however, by the small trial count, the absence of confidence intervals, the lack of a quantitative occlusion analysis, and the paper's own admission that the most failure-prone move (X) relies on a non-topological polygon-centroid fallback when tracking fails after RII. These issues do not invalidate the contribution, but they do mean the central claims as stated are stronger than the evidence currently supports.

major comments (4)
  1. [Section III (X move description)] The manuscript states that tracking often fails to accurately initialize the crossing topology after RII due to depth-resolution sensitivity, and therefore the undertip grasp point S(s_cb,tip) is computed as the nearest tracked node to the centroid of a convex polygon formed from the tracked points, rather than from the topological relation. Since the X move has the lowest per-move success rate (0.615), and the total success rate is exactly the product of the per-move rates (0.937 × 0.867 × 0.615 = 0.500), this fallback is load-bearing for the headline result. The paper should report how many of the 16 trials triggered the fallback, the success rate conditional on fallback versus topological initialization, and how the fallback affects the claim that topological DLO state tracking is the enabling input for the planner.
  2. [Section IV, Table I] The headline 50% success rate is based on 16 trials, but no confidence intervals or exact counts are reported. The implied counts are RI 15/16, RII 13/15, X 8/13, and total 8/16; the 95% confidence interval for 8/16 is roughly [0.28, 0.72], so the difference from the DDOD baseline of 0.66 is not statistically meaningful as reported. Please report exact trial counts for each move, binomial confidence intervals, and a statement of how failures were attributed to perception, planning, or execution, including whether a failed move prevented later moves from being attempted.
  3. [Section IV, Table II and Introduction] The comparison with DDOD, GSP, and Imitation is presented as a table of overall success rates without stating the experimental conditions for each baseline. Knot-tying success depends strongly on the robot, DLO material and length, initial configuration distribution, camera setup, and evaluation protocol, so a bare table of rates does not support a competitive claim such as 'compared to the success rate of the learning-based state-of-the-art method of 66%.' Please specify the setup for each baseline, or explicitly reframe the table as an informal reference point rather than a controlled comparison.
  4. [Abstract and Section IV (claims of robustness and repeatability)] The abstract claims the method is 'robust to occlusion' and 'repeatable for varying rope initial configurations,' but the experiments do not measure occlusion severity or systematically vary initial configurations beyond starting from symmetric curved shapes with one taped tip. The paper should define the range of tested variation (e.g., initial curvature, camera viewpoint, self-occlusion duration) and report the failure breakdown by cause. Without this, the abstract's robustness and repeatability claims exceed the evidence presented.
minor comments (5)
  1. [Equation (3)] The typeset equation for the geodesic distance is difficult to parse; in particular, the summation in the second case is not clearly rendered. Please check the equation formatting so that the piecewise definition is unambiguous.
  2. [Section IV, final paragraph] The sentence 'Manipulation assumes the perceived topology of the DLO is accurate before performing RI and RII, but not before performing RIII' refers to RIII, but the executed sequence is RI, RII, and X; RIII is never performed. This should be corrected to refer to the X move or removed.
  3. [Figures 2 and 5] The figure captions are too brief to identify the seven curvilinear lengths (s=0.5, s=λ, s=1−λ, s_cb, s_ct, s_cb,tip, s_ct,tip) on the depicted rope. Annotating a representative image would substantially improve interpretability, which is a central claimed advantage.
  4. [Section IV, parameter values] The hand-selected constants λ=0.1, γ=0.4, r=5, and M=30 appear only in the experiments section; a sentence explaining how these values were chosen, or a sensitivity study, would clarify the claim of independence from task parameterizations.
  5. [Section IV, text] There is a typo in 'tying an an overhand knot'; also, the abbreviation DDOD in Table II is not defined at first use in the text.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the 50% success rate is a measured experimental outcome, and the waypoint planner is not a re-statement of its inputs.

full rationale

KnotDLO's derivation chain is: TrackDLO estimates the DLO curve S; the planner indexes S at selected curvilinear lengths (such as s=0.5, lambda, and 1-lambda) and computes grasp and target waypoints from tangents and geodesic distances; the robot executes RI, RII, and X; success is scored physically. None of these steps fits a parameter to the outcome and then calls it a prediction. The constants lambda, gamma, and r are fixed hand-set parameters, not fit to the reported 50% success rate. The overall success rate is the measured product of per-move successes (0.937 x 0.867 x 0.615 = 0.500), not a derived identity. The paper does rely on the authors' own TrackDLO tracker as perception input, but TrackDLO is a previously published, code-released component, and the paper transparently reports its failure mode, acknowledging that tracking often fails to initialize crossing topology after RII. That admission is a robustness limitation, not circularity. The replacement of the topological undertip computation S(scb,tip) with a polygon-centroid heuristic for the X move weakens the 'topology-driven' characterization of that move, but the X move's success rate is still measured experimentally, not implied by the centroid definition. No equation or construction makes an output equal to an input by definition. The central claim is therefore self-contained with respect to its experimental evaluation, and any weaknesses are correctness or generalization concerns rather than circular reasoning.

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

The central claim rests on four hand-chosen constants and four assumptions about knot theory, scene setup, and tracking accuracy. No new physical entities are introduced. The system's own text admits the tracking-accuracy assumption is violated after RII, which is the load-bearing fragility.

free parameters (4)
  • lambda = 0.1
    Curvilinear-length offset for the s=lambda and s=1-lambda waypoints; hand-selected for the 0.88 m rope.
  • gamma = 0.4
    Scales lift height h for RI and RII via h = gamma * rho(...); chosen by hand.
  • r = 5
    Node radius around the undertip for selecting the X-move grasp; chosen by hand.
  • M = 30
    Number of control points in the TrackDLO curve; a tracking parameter adopted without sensitivity analysis.
assumptions (4)
  • standard math Reidemeister moves transform a DLO projection between knot states and RI+RII+X compose an overhand knot.
    Invoked in Section II with citation to Reidemeister [25]. This is accepted knot theory.
  • domain assumption The DLO is manipulated semi-planarly with z = [0,0,1], with grasp orientation derived from the curve tangent.
    Section III defines the semi-planar manipulation constraint and all waypoint orientations assume it.
  • domain assumption The DLO starts in a symmetric curved shape and can be distinguished from the background by HSV thresholding, with one tip marked by green tape.
    Section IV states this as the experimental setup; it bounds the claim of previously unseen configurations.
  • domain assumption TrackDLO's perceived topology is accurate before RI and RII.
    Section IV explicitly states that manipulation assumes the perceived topology is accurate before RI and RII, but not before RIII. The X move is made robust to tracking failure via a polygon centroid heuristic.

how reviews work

0 comments
Cite this review

Pith. "Pith review of KnotDLO: Toward Interpretable Knot Tying." pith.science (2026). https://pith.science/paper/WJ5GB3EZ

@misc{pith2026250622176,
  author       = {Pith},
  title        = {Pith review of: KnotDLO: Toward Interpretable Knot Tying},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WJ5GB3EZ}},
  note         = {Machine review of arXiv:2506.22176}
}
read the original abstract

This work presents KnotDLO, a method for one-handed Deformable Linear Object (DLO) knot tying that is robust to occlusion, repeatable for varying rope initial configurations, interpretable for generating motion policies, and requires no human demonstrations or training. Grasp and target waypoints for future DLO states are planned from the current DLO shape. Grasp poses are computed from indexing the tracked piecewise linear curve representing the DLO state based on the current curve shape and are piecewise continuous. KnotDLO computes intermediate waypoints from the geometry of the current DLO state and the desired next state. The system decouples visual reasoning from control. In 16 trials of knot tying, KnotDLO achieves a 50% success rate in tying an overhand knot from previously unseen configurations.

Figures

Figures reproduced from arXiv: 2506.22176 by the authors.

Figure 1
Figure 1. One robot arm with a parallel gripper grasps the DLO and moves it [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Given an initial symmetric, curve-shaped DLO configuration, the planning system computes grasp and target indices from the topology to compose [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Four movement primitives can transform a DLO between any knot [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: A grasp pose can be computed given the curve representation of [PITH_FULL_IMAGE:figures/full_fig_p003_4.png]
Figure 5
Figure 5. Figure 5: A small number of waypoints designed based on the rope topology [PITH_FULL_IMAGE:figures/full_fig_p003_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 35 canonical work pages

  1. [1]

    The Robustness of Tether Friction in Non-Idealized Terrains,

    J. J. Page, L. K. Treers, S. Jorgensen, R. Fearing, and H. S. Stuart, “The Robustness of Tether Friction in Non-Idealized Terrains,” IEEE Robot. Autom. Lett. , vol. 8, no. 1, pp. 424–431, 2023. 1

    work page 2023

  2. [2]

    Friction Binding Study and Remedy Design for Tethered Search and Rescue Robots,

    A. Specian and M. Yim, “Friction Binding Study and Remedy Design for Tethered Search and Rescue Robots,” in IEEE Int. Symp. on Safety, Security, and Rescue Robotics (SSRR) , 2015, pp. 1–6. 1

    work page 2015

  3. [3]

    A System for Robotic Heart Surgery that Learns to Tie Knots Using Recurrent Neural Networks,

    H. Mayer, F. Gomez, D. Wierstra, I. Nagy, A. Knoll, and J. Schmid- huber, “A System for Robotic Heart Surgery that Learns to Tie Knots Using Recurrent Neural Networks,” in IEEE/RSJ Int. Conf. Intell. Robot. Sys. (IROS) , 2006, pp. 543–548. 1

    work page 2006

  4. [4]

    A Novel Knot-Tying Approach for Minimally Invasive Surgical Robot Systems,

    S. Wang, H. Wang, and L. Yue, “A Novel Knot-Tying Approach for Minimally Invasive Surgical Robot Systems,” Int. J. Med. Robot. Comput. Assist. Surg. , vol. 4, no. 3, pp. 268–276, 2008. 1

    work page 2008

  5. [5]

    Suturing and Tying Knots Assisted by a Surgical Robot System in Laryngeal MIS,

    H. Wang, S. Wang, J. Ding, and H. Luo, “Suturing and Tying Knots Assisted by a Surgical Robot System in Laryngeal MIS,” Robotica, vol. 28, no. 2, pp. 241–252, 2010. 1

    work page 2010

  6. [6]

    Improved Knot-Tying Methods for Autonomous Robot Surgery,

    D.-L. Chow and W. Newman, “Improved Knot-Tying Methods for Autonomous Robot Surgery,” in IEEE Int. Conf. Autom. Sci. Eng. (CASE), 2013, pp. 461–465. 1

    work page 2013

  7. [7]

    Disentangling Dense Multi-Cable Knots,

    V . Viswanath, J. Grannen, P. Sundaresan, B. Thananjeyan, A. Balakr- ishna, E. Novoseller, J. Ichnowski, M. Laskey, J. E. Gonzalez, and K. Goldberg, “Disentangling Dense Multi-Cable Knots,” in IEEE/RSJ Int. Conf. Intell. Robot. Sys. (IROS) , 2021, pp. 3731–3738. 1

    work page 2021

  8. [8]

    Learning Where to Trust Unreliable Models in an Unstructured World for Deformable Object Manipulation,

    P. Mitrano, D. McConachie, and D. Berenson, “Learning Where to Trust Unreliable Models in an Unstructured World for Deformable Object Manipulation,” Sci. Rob., vol. 6, no. 54, pp. 1–12, 2021. 1

    work page 2021

Show all 36 references
  1. [9]

    Vision-Based Robotic Grasping and Manipulation of USB Wires,

    X. Li, X. Su, Y . Gao, and Y .-H. Liu, “Vision-Based Robotic Grasping and Manipulation of USB Wires,” in IEEE Int. Conf. Robot. Autom. (ICRA), 2018, pp. 3482–3487. 1

    2018

  2. [10]

    Trajectory Optimization for Manipulation of Deformable Objects: Assembly of Belt Drive Units,

    S. Jin, D. Romeres, A. Ragunathan, D. K. Jha, and M. Tomizuka, “Trajectory Optimization for Manipulation of Deformable Objects: Assembly of Belt Drive Units,” in IEEE Int. Conf. Robot. Autom. (ICRA), 2021, pp. 10 002–10 008. 1

    2021

  3. [11]

    An Online Method for Tight-Tolerance Insertion Tasks for String and Rope,

    W. Wang, D. Berenson, and D. Balkcom, “An Online Method for Tight-Tolerance Insertion Tasks for String and Rope,” in IEEE Int. Conf. Robot. Autom. (ICRA) , 2015, pp. 2488–2495. 1

    2015

  4. [12]

    Knotting/Unknotting Manip- ulation of Deformable Linear Objects,

    H. Wakamatsu, E. Arai, and S. Hirai, “Knotting/Unknotting Manip- ulation of Deformable Linear Objects,” Int. J. Robot. Res. , vol. 25, no. 4, pp. 371–395, 2006. 1

    2006

  5. [13]

    Autonomous Robotic Manipulation of Deformable Linear Objects During Deep Space Maintenance and Repair Procedures,

    D. Rojas, “Autonomous Robotic Manipulation of Deformable Linear Objects During Deep Space Maintenance and Repair Procedures,” Ph.D. dissertation, UC Davis, 2022. 1

    2022

  6. [14]

    MultiDLO: Simultaneous Shape Tracking of Multiple Deformable Linear Objects with Global-Local Topology Preservation,

    J. Xiang and H. Dinkel, “MultiDLO: Simultaneous Shape Tracking of Multiple Deformable Linear Objects with Global-Local Topology Preservation,” in IEEE Int. Conf. Robot. Autom. (ICRA) Workshop on Representing and Manipulating Deformable Objects , 5 2023. 1

    2023

  7. [15]

    Learning Rope Manipulation Policies Using Dense Object Descriptors Trained on Synthetic Depth Data,

    P. Sundaresan, J. Grannen, B. Thananjeyan, A. Balakrishna, M. Laskey, K. Stone, J. E. Gonzalez, and K. Goldberg, “Learning Rope Manipulation Policies Using Dense Object Descriptors Trained on Synthetic Depth Data,” in IEEE Int. Conf. Robot. Autom. (ICRA) , 2020, pp. 9411–9418. 1, 3

    2020

  8. [16]

    Self-Supervised Learning of State Estimation for Manipulating Deformable Linear Objects,

    M. Yan, Y . Zhu, N. Jin, and J. Bohg, “Self-Supervised Learning of State Estimation for Manipulating Deformable Linear Objects,” in IEEE Robot. Autom. Lett. , vol. 5, no. 2, 4 2020, pp. 2372–2379. 1

    2020

  9. [17]

    Learning Topological Motion Primitives for Knot Planning,

    M. Yan, G. Li, Y . Zhu, and J. Bohg, “Learning Topological Motion Primitives for Knot Planning,” in IEEE/RSJ Int. Conf. Intell. Robot. Sys. (IROS), 2020. 1

    2020

  10. [18]

    Zero-Shot Visual Imitation,

    D. Pathak, P. Mahmoudieh, G. Luo, P. Agrawal, D. Chen, Y . Shentu, E. Shelhamer, J. Malik, A. A. Efros, and T. Darrell, “Zero-Shot Visual Imitation,” in Int. Conf. Learn. Represent. , 2018. 1, 3

    2018

  11. [19]

    Combining Self-Supervised Learning and Imitation for Vision-Based Rope Manipulation,

    A. Nair, D. Chen, P. Agrawal, P. Isola, P. Abbeel, J. Malik, and S. Levine, “Combining Self-Supervised Learning and Imitation for Vision-Based Rope Manipulation,” in IEEE Int. Conf. Robot. Autom. (ICRA), 2017, pp. 2146–2153. 1, 3

    2017

  12. [20]

    Shape Control of Deformable Linear Objects with Offline and Online Learning of Local Linear Deformation Models,

    M. Yu, H. Zhong, and X. Li, “Shape Control of Deformable Linear Objects with Offline and Online Learning of Local Linear Deformation Models,” in IEEE Int. Conf. Robot. Autom. (ICRA) , 5 2022, pp. 1337–

    2022

  13. [21]

    Global Model Learning for Large Deformation Control of Elastic Deformable Linear Objects: An Efficient and Adaptive Approach,

    M. Yu, K. Lv, H. Zhong, S. Song, and X. Li, “Global Model Learning for Large Deformation Control of Elastic Deformable Linear Objects: An Efficient and Adaptive Approach,” IEEE Trans. Robot. , vol. 39, no. 1, pp. 417–436, 2023. 1

    2023

  14. [22]

    Ariadne+: Deep Learning–Based Augmented Framework for the Instance Segmentation of Wires,

    A. Caporali, R. Zanella, D. D. Greogrio, and G. Palli, “Ariadne+: Deep Learning–Based Augmented Framework for the Instance Segmentation of Wires,” IEEE Trans. Ind. Informat. , vol. 18, no. 12, pp. 8607–8617,

  15. [23]

    Tracking Partially- Occluded Deformable Objects while Enforcing Geometric Con- straints,

    Y . Wang, D. M cConachie, and D. Berenson, “Tracking Partially- Occluded Deformable Objects while Enforcing Geometric Con- straints,” in IEEE Int. Conf. Robot. Autom. (ICRA) , 2021, pp. 14 199– 14 205. 1

    2021

  16. [24]

    TrackDLO: Tracking Deformable Linear Objects Under Occlusion With Motion Coherence,

    J. Xiang, H. Dinkel, H. Zhao, N. Gao, B. Coltin, T. Smith, and T. Bretl, “TrackDLO: Tracking Deformable Linear Objects Under Occlusion With Motion Coherence,” IEEE Robot. Autom. Lett. , vol. 8, no. 10, pp. 6179–6186, 2023. 1, 2, 3

    2023

  17. [25]

    Reidemeister, Knot Theory

    K. Reidemeister, Knot Theory . Springer-Verlag, 1932. 2

    1932

  18. [26]

    Shapely,

    S. Gillies, C. van der Wel, J. Van den Bossche, M. W. Taves, J. Arnott, B. C. Ward et al. , “Shapely,” 10 2022. [Online]. Available: https://github.com/shapely/shapely 3

    2022

  19. [27]

    Learning to Rearrange Deformable Cables, Fabrics, and Bags with Goal-Conditioned Transporter Networks,

    D. Seita, P. Florence, J. Tompson, E. Coumans, V . Sindhwani, K. Gold- berg, and A. Zeng, “Learning to Rearrange Deformable Cables, Fabrics, and Bags with Goal-Conditioned Transporter Networks,” in IEEE Int. Conf. Robot. Autom. (ICRA) , 2021, pp. 4568–4575. 3

    2021

  20. [28]

    Robotic Operating Sys- tem: Noetic Ninjemys,

    Stanford Artificial Intelligence Laboratory, “Robotic Operating Sys- tem: Noetic Ninjemys,” https://www.ros.org, 2018. 3

    2018

  21. [29]

    The OpenCV Library,

    G. Bradski, “The OpenCV Library,” Dr . Dobb’s Journal of Software Tools, 2000. 3

    2000

  22. [30]

    Array Programming with NumPy,

    C. R. Harris, J. Millman, S. van der Walt, R. Gommers, P. Virtanen, D. Cournapeau, E. Wieser, J. Taylor, S. Berg, N. J. Smith, R. Kern, M. Picus, S. Hoyer, M. H. van Kerkwijk, M. Brett, A. Haldane, J. Fern ´andez del R ´ıo, M. Wiebe, P. Peterson, P. G ´erard-Marchant, K. Shepp...

    2020

  23. [31]

    Matplotlib: A 2D Graphics Environment,

    J. D. Hunter, “Matplotlib: A 2D Graphics Environment,” Comput. Sci. Eng., vol. 9, no. 3, pp. 90–95, 2007. 3

    2007

  24. [32]

    Open3D: A Modern Library for 3D Data Processing,

    Q.-Y . Zhou, J. Park, and V . Koltun, “Open3D: A Modern Library for 3D Data Processing,” arXiv:1801.09847, 2018. 3

    2018 arXiv

  25. [33]

    Scipy 1.0: Fundamental Algorithms for Scientific Computing in Python,

    P. Virtan, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cournapeau, E. Burovski, P. Peterson, W. Weckesser, J. Bright et al. , “Scipy 1.0: Fundamental Algorithms for Scientific Computing in Python,” Nat. Methods , vol. 17, pp. 261—-272, 2020. 3

    2020

  26. [34]

    General Hand–Eye Calibration Based on Reprojection Error Minimization,

    K. Koide and E. Menegatti, “General Hand–Eye Calibration Based on Reprojection Error Minimization,” IEEE Robot. Autom. Lett. , vol. 4, no. 2, pp. 1021–1028, 2019. 3

    2019

  27. [35]

    Eigen v3,

    G. Guennebaud, B. Jacob et al., “Eigen v3,” http://eigen.tuxfamily.org,

  28. [36]

    3D is Here: Point Cloud Library (PCL),

    R. B. Rusu and S. Cousins, “3D is Here: Point Cloud Library (PCL),” in IEEE Int. Conf. Robot. Autom. (ICRA) , 2011, pp. 1–4. 3

    2011

Pith tools

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