Pith. sign in

REVIEW 4 major objections 5 minor 33 references

The paper claims that a feedback loop which stretches tissue after each cut and plans corrective dissection goals lets autonomous surgical robots detect and fix incomplete cuts, raising success rates from 0–45.5% to 80%.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 11:30 UTC pith:LKWRTKWF

load-bearing objection The feedback loop idea is genuinely useful and the robot results are encouraging, but the unreported τ threshold and a somewhat sloppy success metric mean the headline numbers should be treated as promising rather than proven. the 4 major comments →

arxiv 2510.04074 v2 pith:LKWRTKWF submitted 2025-10-05 cs.RO

Feedback Matters: Augmenting Autonomous Dissection with Visual and Topological Feedback

Chung-Pang Wang , Changwei Chen , Xiao Liang , Soofiyan Atar , Florian Richter , Michael Yip This is my paper
classification cs.RO
keywords autonomous dissectionsurgical roboticstissue connectivity estimationfeedback controlexposure maximizationkeypoint trackingrecovery planningdeformable tissue
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper argues that autonomous tissue dissection fails mainly because the robot cannot tell whether a cut actually severed everything it was supposed to sever. Its solution is a closed feedback loop: after each cutting attempt, the robot actively stretches and reorients the tissue so the camera can see how the cut edges deform, infers which connections remain by measuring how much each edge elongated, and then plans a new corrective cut for the remaining attachments. The authors show that adding this feedback raises the success rate of two very different dissection agents — one planning-based, one learned — from 0–45.5% to 80% on a phantom-tissue task. The central claim is that topological state (connected vs severed) can be read reliably from image-space keypoint motion, provided the robot is willing to manipulate the tissue to make that motion visible.

Core claim

The central claim is that post-dissection tissue connectivity can be estimated from image keypoint elongation: after the robot retracts the tissue, an edge that has been severed stretches to at least τ times its original length (ρ ≥ τ) while a still-connected edge does not, and this rule lets the system classify edges as connected or disconnected accurately (86% with the exposure controller, vs 5.56% without). On top of this estimate, a recovery planner treats the uncut edges as a set-cover instance and greedily proposes new dissection goals, and the loop repeats (up to three attempts) until no uncut edge intersects the goal. With this feedback, both a planning-based and a learned dissection

What carries the argument

The load-bearing object is the elongation ratio ρ(e_ij) = ||k_i(t_e)−k_j(t_e)|| / ||k_i(t_d)−k_j(t_d)|| between pairs of tracked keypoints sampled on a grid around the dissection goal: an edge is declared dissected when ρ ≥ τ, with τ > 1 a fixed threshold that the paper never states. Reliable measurement of ρ is ensured by an exposure-maximization controller, which solves a finite-horizon optimal control problem (MPPI on an XPBD tissue model) with a camera-facing normal loss, a deformation-expanding loss aligned with the dissection goal's normal, and a visibility constraint, so that the tissue is stretched orthogonally to the cut and rotated toward the camera. The recovery planner then conve

Load-bearing premise

The entire feedback loop rests on the rule that severed tissue edges stretch to at least τ times their original length while connected edges do not (Eqs. 5–6), and the paper never reports the value of τ or tests how sensitive the results are to it.

What would settle it

Run the same phantom-tissue experiment but deliberately pick τ values across a range (say 1.05, 1.2, 1.5, 2.0) and record connectivity-estimation accuracy and final dissection success for each; if success collapses for any τ near the implied operating point, or if connected edges routinely exceed the threshold under the exposure controller, the feedback estimate is not reliable.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If correct, autonomous dissection can close the loop using only the existing endoscopic camera and keypoint tracking; no new sensors are required to detect incomplete cuts.
  • The feedback loop is policy-agnostic: it improved both a learning-based and a planning-based agent, so it can be added on top of other dissection controllers.
  • Corrective goals are generated as small, localized segments, which should reduce excessive cutting over repeated attempts (the effective cut ratio rises and excessive cut ratio falls across the three attempts).
  • The framework demonstrates a general principle: active manipulation to maximize visibility is a form of perception, not just execution.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The τ threshold is never reported, so the method's practical reliability rests on an unstated calibration step; a sensitivity analysis against τ would test whether the 80% result is robust.
  • The connectivity rule assumes 2D image-space elongation faithfully tracks 3D topology during stretching; under oblique camera angles or large out-of-plane rotations this could break, suggesting a testable extension using multi-view or depth-aware verification.
  • Because the method targets thin-shell tissues (stated limitation), extending to volumetric dissection would require tracking interior structure, perhaps via ultrasound or pre-operative imaging — a natural next step the authors flag.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The manuscript presents a closed-loop feedback framework for autonomous tissue dissection on the da Vinci Research Kit. After a dissection action, tissue connectivity is estimated from the elongation of image keypoints on a grid around the dissection goal: an edge is declared cut if its tracked length ratio exceeds a threshold τ (Eqs. 5–6). An exposure-maximization controller (MPPI with camera-facing and deformation losses) actively retracts tissue to make this estimate reliable. A recovery planner then solves a set-cover problem (Algorithm 1) to propose corrective dissection goals. Experiments evaluate (i) detection accuracy with three keypoint trackers on beef, chicken, and silicone; (ii) visible area and estimation accuracy with/without the exposure controller; (iii) full-pipeline success of planning- and learning-based dissection agents with and without feedback on a silicone phantom; and (iv) one real chicken-tissue demonstration. The central claim is that feedback substantially improves success rates (80% vs. 0–45.5%) and reduces errors.

Significance. If the reported performance is correct, the paper makes a tangible contribution: it identifies a practical signal—image-keypoint elongation under active tissue manipulation—for verifying dissection completeness, and it demonstrates that this signal can be integrated into two very different dissection policies. The component evaluations have clear non-circular checks: detection accuracy is compared with human labels, and visible area is measured by manually contoured regions. The comparison against no-feedback and repeated-no-feedback baselines is the right experimental design. However, the load-bearing threshold τ and the internal consistency of Table III need to be resolved before the central claim can be accepted.

major comments (4)
  1. [§III-B, Eqs. (5)–(6)] The connectivity estimator reduces to a single elongation threshold τ, which is never reported. Every downstream object—E_uncut in Eq. (12), the set cover in Eq. (13), and the corrective goals from Algorithm 1—inherits this threshold. The exposure controller in Eq. (10) deliberately stretches tissue along the dissection normal, so connected tissue can also exceed any fixed τ, while camera foreshortening or partial cuts can keep severed edges below τ. The detection accuracies in Table I were computed with the same hidden threshold and do not bound the false-positive/false-negative trade-off. Please report τ, provide precision/recall or an ROC-style curve over τ for the curated 50-case dataset, and evaluate sensitivity across materials, deformation magnitudes, and camera viewpoints.
  2. [§IV-C, Table III] The table does not make the units of 'Remaining Attachments' explicit. Under the stated success criterion (zero Remaining Attachments), the success rates in rows with large positive attachment counts are only interpretable if this column is a total across all ten trials, not a per-trial average; with an average of 16, for example, a 20% success rate would require a very skewed distribution. Moreover, planning-based w/o feedback (n=3) shows lower mean Length Deviation and lower attachments than n=1, yet Success Rate drops from 20% to 0%, which is unexplained. Please state units and report raw per-trial values or a scatter of success/failure.
  3. [§IV-A.2] The subsection 'Quality of Recovery Feedback Planner' contains no quantitative result. The claim that the recovery planner is robust and generalizes across tissues, dissection states, and tools is supported only by the qualitative Fig. 3. Because the recovery planner is the component that turns E_uncut into corrective goals (Algorithm 1), its standalone evaluation should include, e.g., the fraction of ground-truth uncut edges covered by G_next, the number of corrective goals, false-positive corrective cuts, and comparison with a baseline planner. Without this, the full-pipeline success in Table III cannot be attributed to the recovery planner as opposed to the estimator/controller.
  4. [§III-A, §IV-C] The learning-based dissection policy is under-specified: no network architecture, observation/action parameterization, number of demonstrations, or training procedure is given. Since Table III reports learning-based success rates and the paper claims versatility across dissection paradigms, this policy must be described or referenced to an existing method. Otherwise the learning-based results cannot be reproduced or compared with prior work.
minor comments (5)
  1. [§III-C, Eq. (7)/(10)] Values for the MPPI horizon h, cost weights λ_c and λ_d, number of samples, and the candidate distance threshold d̂ in Eq. (14) are not reported. Please add an implementation-parameters table.
  2. [§IV-A.1, Table I] Detection accuracy is reported only as an average across all materials and conditions. Since beef is feature-rich and chicken skin is feature-poor, a per-material breakdown would strengthen the claim of generalizability.
  3. [§IV-B] For the exposure-controller experiment, please clarify whether the 20 trials are the same dataset used in §IV-A, and state the tissue type, initial dissection state, and how the ground-truth connectivity labels were obtained. The 69 px visible area of the 'naive normal' baseline is very close to occlusion; a sentence explaining why this is expected would help.
  4. [§IV-C, Fig. 5] The figure axes are hard to parse in the text: 'Remaining Attachments' appears with inconsistent units ('#', 'out of 10'), and the text refers to 'Effective Cut Ratio (True Positive)' but the body defines only Length Deviation and Remaining Attachments. Please align axis labels with the definitions in the text.
  5. [Throughout] Minor language and typographical issues: 'dpeth' (Sec. III-A), 'meachnism' (Sec. IV-C), 'it’s reliability' (Sec. I), and inconsistent use of 'OR' in the Fig. 2 caption. Also, the limitation paragraph in Sec. V is good but a short discussion of observed failure modes of the feedback loop (e.g., missed E_uncut due to threshold, incorrect corrective goals) would be valuable.

Circularity Check

0 steps flagged

No significant circularity: component evaluations use independent human labels and the controller/recovery planner are optimizations over stated inputs.

full rationale

The paper's derivation chain is self-contained and not circular. The connectivity estimator is an explicit rule on tracked elongation (Eqs. 5–6) and is evaluated against human ground-truth labels in Sec. IV.A: 'N_correct is the number of pairs whose predicted separation state matches the ground-truth (connected or separated) labeled by human.' The exposure controller (Eqs. 7–11) is an a priori optimal-control formulation with stated costs; its benefit is separately assessed by manually contoured visible area and by connectivity accuracy against human-labeled N_GT (Sec. IV.B). The recovery planner (Eq. 13, Alg. 1) is a set-cover optimization over the estimator's output, not a fitted predictor of final success; success is independently defined in Sec. IV.C as Length Deviation less than 2 mm and zero Remaining Attachments. The unreported threshold τ and the absence of sensitivity analysis are genuine robustness/calibration gaps, but they do not make any result equivalent to its input by construction because the component experiments use external labels rather than outputs of the same pipeline. Self-citations (e.g., [3]) appear as background for XPBD/previous manipulation work and do not carry the central claim.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The central claim rests on the keypoint-elongation rule for detecting uncut tissue, a threshold parameter τ never specified, and on a simulation-based controller whose real-world validity is only empirically sampled. The paper introduces no new physical entities. Several experimental hyperparameters (λc, λd, h, d̂, max attempts) are also unstated, making the method harder to reproduce.

free parameters (5)
  • Elongation threshold τ = not reported (only τ > 1)
    Used in Eq. (5)-(6) to classify each tracked edge as connected or cut; all connectivity estimates and the recovery planner depend on it.
  • MPPI cost weights λc, λd = not reported
    Balance the camera-facing loss and the deformation-expanding loss in Eq. (7); the exposure controller's behavior depends on them.
  • MPPI horizon h and sampling parameters = not reported
    The controller samples future control trajectories with horizon h; no values are given.
  • Candidate distance threshold d̂ and rotation bin size = not reported (bin size 6)
    Defines the candidate set S in Eq. (14) for recovery planning.
  • Success length-deviation threshold = 2 mm
    Arbitrary threshold used in the definition of a successful cut in Sec. IV-C.
axioms (5)
  • domain assumption Severed tissue edges elongate by at least factor τ while connected edges do not, under tissue stretching.
    Linear elongation ratio rule in Eqs. (5)-(6); this is the core of connectivity estimation and of the whole feedback loop.
  • domain assumption XPBD simulation f_xpbd of the reconstructed mesh M_td predicts real tissue deformation well enough for MPPI-chosen actions to transfer to the physical robot.
    The exposure maximization controller in Eq. (7) is solved in simulation; no formal sim-to-real guarantee is provided.
  • domain assumption The dissection goal is a low-curvature polyline in pixel space, and tissue connectivity can be assessed from 2D image keypoints without resolving 3D topology.
    Stated in Sec. III and acknowledged in the Discussion as limiting the method to thin-shell surface tissue.
  • domain assumption The visibility constraint C_vis, checking only that pixels remain inside the image bounds, is sufficient for reliable connectivity estimation.
    Eq. (11) ignores occlusion by instruments, specularities, or poor lighting, which can all affect tracking quality.
  • standard math Greedy selection gives a good approximate solution to the NP-hard minimum set cover problem.
    Algorithm 1 applies the standard greedy set cover approximation; this is a standard result and not a fragile assumption.

pith-pipeline@v1.3.0-alltime-deepseek · 10902 in / 9471 out tokens · 72865 ms · 2026-08-04T11:30:53.975692+00:00 · methodology

0 comments
read the original abstract

Autonomous surgical systems must adapt to highly dynamic environments where tissue properties and visual cues evolve rapidly. Central to such adaptability is feedback: the ability to sense, interpret, and respond to changes during execution. While feedback mechanisms have been explored in surgical robotics, ranging from tool and tissue tracking to error detection, existing methods remain limited in handling the topological and perceptual challenges of tissue dissection. In this work, we propose a feedback-enabled framework for autonomous tissue dissection that explicitly reasons about topological changes from endoscopic images after each dissection action. This structured feedback guides subsequent actions, enabling the system to localize dissection progress and adapt policies online. To improve the reliability of such feedback, we introduce visibility metrics that quantify tissue exposure and formulate optimal controller designs that actively manipulate tissue to maximize visibility. Finally, we integrate these feedback mechanisms with both planning-based and learning-based dissection methods, and demonstrate experimentally that they significantly enhance autonomy, reduce errors, and improve robustness in complex surgical scenarios.

Figures

Figures reproduced from arXiv: 2510.04074 by Changwei Chen, Chung-Pang Wang, Florian Richter, Michael Yip, Soofiyan Atar, Xiao Liang.

Figure 1
Figure 1. Figure 1: Illustration of the importance of exposure maximiza [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Our Autonomous Dissection Pipeline. Our framework operates in a feedback loop. The process begins with a human-provided dissection goal Ghuman, which the autonomous agent executes. Next, an exposure maximization controller manipulates the tissue to improve visibility for the subsequent error estimation. Finally, a recovery planner estimates the remaining tissue connectivity. If the dissection is incomplete… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative results of the recovery feedback planner [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of dissection performance over three attempts: (a) remaining attachments, (b) effective cut ratio, and (c) [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: This figure illustrates the execution of our complete feedback-enabled dissection pipeline. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

33 extracted references · 11 linked inside Pith

  1. [1]

    Differentiable rendering-based pose estimation for surgical robotic instruments,

    Z. Liang, Z.-Y . Chiu, F. Richter, and M. C. Yip, “Differentiable rendering-based pose estimation for surgical robotic instruments,”

  2. [2]

    Robust surgical tool tracking with pixel- based probabilities for projected geometric primitives,

    C. D’Ambrosia, F. Richter, Z.-Y . Chiu, N. Shinde, F. Liu, H. I. Christensen, and M. C. Yip, “Robust surgical tool tracking with pixel- based probabilities for projected geometric primitives,” in2024 IEEE International Conference on Robotics and Automation (ICRA), 2024, pp. 15 455–15 462

    2024

  3. [3]

    Medic: Autonomous surgical robotic assistance to maximizing expo- sure for dissection and cautery,

    X. Liang, C.-P. Wang, N. U. Shinde, F. Liu, F. Richter, and M. Yip, “Medic: Autonomous surgical robotic assistance to maximizing expo- sure for dissection and cautery,” in2025 IEEE International Confer- ence on Robotics and Automation (ICRA), 2025, pp. 6889–6895

    2025

  4. [4]

    Real-time constrained 6d object-pose tracking of an in-hand suture needle for minimally invasive robotic surgery,

    Z.-Y . Chiu, F. Richter, and M. C. Yip, “Real-time constrained 6d object-pose tracking of an in-hand suture needle for minimally invasive robotic surgery,”arXiv preprint arXiv:2210.11973, 2022

    Pith/arXiv arXiv 2022

  5. [5]

    Real-to- sim deformable object manipulation: Optimizing physics models with residual mappings for robotic surgery,

    X. Liang, F. Liu, Y . Zhang, Y . Li, S. Lin, and M. Yip, “Real-to- sim deformable object manipulation: Optimizing physics models with residual mappings for robotic surgery,” in2024 IEEE International Conference on Robotics and Automation (ICRA), 2024, pp. 15 471– 15 477

    2024

  6. [6]

    Srt-h: A hierarchical framework for autonomous surgery via language-conditioned imitation learning,

    J. W. Kim, J.-T. Chen, P. Hansen, L. X. Shi, A. Goldenberg, S. Schmidgall, P. M. Scheikl, A. Deguet, B. M. White, D. R. Tsai et al., “Srt-h: A hierarchical framework for autonomous surgery via language-conditioned imitation learning,”Science robotics, vol. 10, no. 104, p. eadt5254, 2025

    2025

  7. [7]

    Jiggle: An active sensing framework for boundary param- eters estimation in deformable surgical environments,

    N. U. Shinde, X. Liang, F. Liu, Y . Zhang, F. Richter, S. Herbert, and M. C. Yip, “Jiggle: An active sensing framework for boundary param- eters estimation in deformable surgical environments,”Conference on Robotics: Science and Systems (RSS), 2024

    2024

  8. [8]

    Medic: Autonomous surgical robotic assistance to maximizing exposure for dissection and cautery,

    X. Liang, C.-P. Wang, N. U. Shinde, F. Liu, F. Richter, and M. Yip, “Medic: Autonomous surgical robotic assistance to maximizing exposure for dissection and cautery,” 2024. [Online]. Available: https://arxiv.org/abs/2409.14287

    Pith/arXiv arXiv 2024

  9. [9]

    Autonomous robotic suction to clear the surgical field for hemostasis using image-based blood flow detection,

    F. Richter, S. Shen, F. Liu, J. Huang, E. K. Funk, R. K. Orosco, and M. C. Yip, “Autonomous robotic suction to clear the surgical field for hemostasis using image-based blood flow detection,”IEEE Robotics and Automation Letters, vol. 6, no. 2, pp. 1383–1390, 2021

    2021

  10. [10]

    Semi-autonomous robotic assistance for gallblad- der retraction in surgery,

    A. Sch ¨ußler, C. Kunz, R. Younis, B. Alt, J. Paik, M. Wagner, and F. Mathis-Ullrich, “Semi-autonomous robotic assistance for gallblad- der retraction in surgery,”IEEE Robotics and Automation Letters, vol. 10, no. 7, pp. 7468–7475, 2025

    2025

  11. [11]

    Surgical robot transformer (srt): Imitation learning for surgical tasks,

    J. W. Kim, T. Z. Zhao, S. Schmidgall, A. Deguet, M. Kobilarov, C. Finn, and A. Krieger, “Surgical robot transformer (srt): Imitation learning for surgical tasks,”arXiv preprint arXiv:2407.12998, 2024

    Pith/arXiv arXiv 2024

  12. [12]

    Supervised autonomous electrosurgery via biocom- patible near-infrared tissue tracking techniques,

    H. Saeidi, J. Ge, M. Kam, J. D. Opfermann, S. Leonard, A. S. Joshi, and A. Krieger, “Supervised autonomous electrosurgery via biocom- patible near-infrared tissue tracking techniques,”IEEE Transactions on Medical Robotics and Bionics, vol. 1, no. 4, pp. 228–236, 2019

    2019

  13. [13]

    Landmark-guided deformable image registration for supervised au- tonomous robotic tumor resection,

    J. Ge, H. Saeidi, J. D. Opfermann, A. S. Joshi, and A. Krieger, “Landmark-guided deformable image registration for supervised au- tonomous robotic tumor resection,” inMedical Image Computing and Computer Assisted Intervention – MICCAI 2019, D. Shen, T. Liu, T. M. Peters, L. H. Staib, C. Essert, S. Zhou, P.-T. Yap, and A. Khan, Eds. Cham: Springer Internat...

    2019

  14. [14]

    A framework for automated dissection along tissue boundary,

    K.-H. Oh, L. Borgioli, M. Zefran, L. Chen, and P. C. Giulianotti, “A framework for automated dissection along tissue boundary,”arXiv preprint arXiv:2310.09669, 2023

    Pith/arXiv arXiv 2023

  15. [15]

    Supervised autonomous electrosurgery for soft tissue resection,

    J. Ge, H. Saeidi, M. Kam, J. Opfermann, and A. Krieger, “Supervised autonomous electrosurgery for soft tissue resection,” in2021 IEEE 21st International Conference on Bioinformatics and Bioengineering (BIBE). IEEE, 2021, pp. 1–7

    2021

  16. [16]

    Autonomous system for tumor resection (astr) - dual-arm robotic midline partial glossectomy,

    J. Ge, M. Kam, J. D. Opfermann, H. Saeidi, S. Leonard, L. J. Mady, M. J. Schnermann, and A. Krieger, “Autonomous system for tumor resection (astr) - dual-arm robotic midline partial glossectomy,”IEEE Robotics and Automation Letters, vol. 9, no. 2, pp. 1166–1173, 2024

    2024

  17. [17]

    Reward learning from suboptimal demonstrations with applications in surgical electrocautery,

    Z. Karimi, S.-H. Ho, B. Thach, A. Kuntz, and D. S. Brown, “Reward learning from suboptimal demonstrations with applications in surgical electrocautery,”arXiv preprint arXiv:2404.07185, 2024

    Pith/arXiv arXiv 2024

  18. [18]

    An ultrasound-guided system for autonomous marking of tumor bound- aries during robot-assisted surgery,

    N. Marahrens, D. Jones, N. Murasovs, C. Biyani, and P. Valdastri, “An ultrasound-guided system for autonomous marking of tumor bound- aries during robot-assisted surgery,”IEEE Transactions on Medical Robotics and Bionics, 2024

    2024

  19. [19]

    Image-based incision detection for topological intraoperative 3d model update in augmented reality assisted laparoscopic surgery,

    T. Franc ¸ois, L. Calvet, C. S `eve-d’Erceville, N. Bourdel, and A. Bar- toli, “Image-based incision detection for topological intraoperative 3d model update in augmented reality assisted laparoscopic surgery,” inMedical Image Computing and Computer Assisted Intervention– MICCAI 2021: 24th International Conference, Strasbourg, France, September 27–October...

    2021

  20. [20]

    Enhancing surgical precision in autonomous robotic incisions via physics-based tissue cutting simulation,

    J. Ge, E. Kilmer, L. J. Mady, J. D. Opfermann, and A. Krieger, “Enhancing surgical precision in autonomous robotic incisions via physics-based tissue cutting simulation,” in2024 IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems (IROS). IEEE, 2024, pp. 2421–2428

    2024

  21. [21]

    Cressim-mpm: A material point method library for surgical soft body simulation with cutting and suturing,

    Y . Ou and M. Tavakoli, “Cressim-mpm: A material point method library for surgical soft body simulation with cutting and suturing,” arXiv preprint arXiv:2502.18437, 2025

    Pith/arXiv arXiv 2025

  22. [22]

    Cotracker: It is better to track together,

    N. Karaev, I. Rocco, B. Graham, N. Neverova, A. Vedaldi, and C. Rupprecht, “Cotracker: It is better to track together,” inProc. ECCV, 2024

    2024

  23. [23]

    Stereo anywhere: Robust zero-shot deep stereo matching even where either stereo or mono fail,

    L. Bartolomei, F. Tosi, M. Poggi, and S. Mattoccia, “Stereo anywhere: Robust zero-shot deep stereo matching even where either stereo or mono fail,” inProceedings of the Computer Vision and Pattern Recognition Conference (CVPR), June 2025, pp. 1013–1027

    2025

  24. [24]

    Video depth anything: Consistent depth estimation for super-long videos,

    S. Chen, H. Guo, S. Zhu, F. Zhang, Z. Huang, J. Feng, and B. Kang, “Video depth anything: Consistent depth estimation for super-long videos,”arXiv:2501.12375, 2025

    Pith/arXiv arXiv 2025

  25. [25]

    Depth anything v2,

    L. Yang, B. Kang, Z. Huang, Z. Zhao, X. Xu, J. Feng, and H. Zhao, “Depth anything v2,”arXiv:2406.09414, 2024

    Pith/arXiv arXiv 2024

  26. [26]

    Real-to-sim de- formable object manipulation: Optimizing physics models with resid- ual mappings for robotic surgery,

    X. Liang, F. Liu, Y . Zhang, Y . Li, S. Lin, and M. Yip, “Real-to-sim de- formable object manipulation: Optimizing physics models with resid- ual mappings for robotic surgery,”arXiv preprint arXiv:2309.11656, 2023

    Pith/arXiv arXiv 2023

  27. [27]

    Model predictive path integral control using covariance variable importance sampling,

    G. Williams, A. Aldrich, and E. Theodorou, “Model predictive path integral control using covariance variable importance sampling,”arXiv preprint arXiv:1509.01149, 2015

    Pith/arXiv arXiv 2015

  28. [28]

    Warp: A high-performance python framework for gpu simulation and graphics,

    M. Macklin, “Warp: A high-performance python framework for gpu simulation and graphics,” https://github.com/nvidia/warp, March 2022, nVIDIA GPU Technology Conference (GTC)

    2022

  29. [29]

    An open-source research kit for the da vinci® surgical system,

    P. Kazanzides, Z. Chen, A. Deguet, G. S. Fischer, R. H. Taylor, and S. P. DiMaio, “An open-source research kit for the da vinci® surgical system,” in2014 IEEE International Conference on Robotics and Automation (ICRA), 2014, pp. 6434–6439

    2014

  30. [30]

    CoTracker3: Simpler and better point tracking by pseudo-labelling real videos,

    N. Karaev, I. Makarov, J. Wang, N. Neverova, A. Vedaldi, and C. Rupprecht, “CoTracker3: Simpler and better point tracking by pseudo-labelling real videos,” 2024

    2024

  31. [31]

    Raft: Recurrent all-pairs field transforms for optical flow,

    Z. Teed and J. Deng, “Raft: Recurrent all-pairs field transforms for optical flow,” 2020. [Online]. Available: https://arxiv.org/abs/2003. 12039

    2020

  32. [32]

    Litetracker: Leveraging temporal causality for accurate low-latency tissue tracking,

    M. A. Karaoglu, W. Ji, A. Abbas, N. Navab, B. Busam, and A. Ladikos, “Litetracker: Leveraging temporal causality for accurate low-latency tissue tracking,” 2025. [Online]. Available: https://arxiv.org/abs/2504.09904

    arXiv 2025

  33. [2025]

    Available: https://arxiv.org/abs/2503.05953

    [Online]. Available: https://arxiv.org/abs/2503.05953

    Pith/arXiv arXiv