Certifiably Safe Manipulation of Deformable Linear Objects via Joint Shape and Tension Prediction

Shichen Li; Yiting Zhang

REVIEW 3 major objections 5 minor 1 cited by

Certifiably Safe Manipulation of Deformable Linear Objects via Joint Shape and Tension Prediction

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A robot can manipulate deformable linear objects safely by predicting both future shape and tension and rejecting any planned trajectory whose worst-case tension exceeds a safety limit.

desk verdict A sensible learning-plus-reachability integration for DLO manipulation that overclaims 'certifiable' safety by setting the error bound to a test-set maximum. read the letter →

arxiv 2505.13889 v1 pith:LWJY7CFI submitted 2025-05-20 cs.RO cs.LG

Yiting Zhang , Shichen Li This is my paper · ORCID

classification cs.ROcs.LG

keywords deformablelinearobjectstensionpredictioncertifiablysafemotionplanningpolynomialzonotopesreachabilityanalysiswireharnessassemblycontact-richmanipulationLSTM

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

The paper claims that a robot can manipulate cables, ropes, and other deformable linear objects safely in cluttered, contact-rich scenes by predicting both the future shape and the internal tension of the object, not just its shape. It builds a planner that evaluates whole families of robot trajectories, using a learned model to forecast DLO shape and tension, and rejects any trajectory whose predicted tension could exceed a safety limit. The authors test this in a simulated wire-harness assembly task: over 100 randomized trials the robot reaches the goal 76 times, never collides, and never overstretches the harness. If the claim holds, tension prediction is the missing ingredient that lets robots exploit necessary contact with the environment without sacrificing safety.

What carries the argument

The central mechanism is the joint shape-and-tension predictor and its integration with polynomial zonotope reachability. Polynomial zonotopes are set representations that overapproximate a family of trajectories parameterized by $k$ as polynomial functions of time and parameters, which lets the optimizer propagate uncertainty through the robot dynamics and the DLO tension model. The learned predictor is an LSTM-based network with a quadratic-programming contact handler, trained on 2,000 simulated trajectories, that produces the state and tension forecasts used in the cost function and safety constraints. The load-bearing identity is the safety constraint $\lVert f(t_i;k)\rVert + \varepsilon \le f_{\text{lim}}$, which turns a predicted tension bound into a worst-case guarantee against overstretching, and Assumption 3 is the premise that such an $\varepsilon$ exists uniformly.

What would settle it

Run the planner in an environment with a force sensor on the fixed end of the DLO, on a set of task configurations and trajectory parameters not used in training; if any trajectory is found where the predicted tension satisfies $\lVert f(t_i;k)\rVert + \varepsilon \le f_{\text{lim}}$ while the measured true tension exceeds $f_{\text{lim}}$, then the certified safety claim is false as stated.

Watch

Extended reading notes

Core claim

The paper's central claim is that DLO manipulation can be made certifiably safe by adding tension prediction to shape prediction inside a reachability-based trajectory optimizer. The learned model takes the current DLO state and the planned end-effector path and outputs both the future DLO shape and the tension profile, with a contact handler that projects penetrated nodes back onto obstacle surfaces. The planner overapproximates the reachable sets of joint configurations, velocities, inputs, occupancy, and DLO tension using polynomial zonotopes over a continuum of trajectory parameters. Safety is enforced by constraints including joint limits, collision avoidance, and the new constraint $\lVert f(t_i;k)\rVert + \varepsilon \le f_{\text{lim}}$, where $\varepsilon$ is an assumed certified bound on tension-prediction error. Under this constraint, even a worst-case underestimate of tension stays below the overstretch limit. In 100 simulated trials of wire-harness assembly, the framework reported a 76% success rate with zero robot collisions and zero DLO overextensions.

Load-bearing premise

The whole safety certificate rests on Assumption 3, that the learned tension model's error is always below epsilon for every time step and every trajectory the planner could choose, but the paper sets epsilon to the largest error on a finite test set, which does not prove a bound over the continuous trajectory family.

Editorial extensions

If this is right

Executing a planned trajectory guarantees, under Assumption 3, no joint-limit violations, no collisions, and no tension above the safety threshold.
Robots can use necessary contact with the environment during DLO manipulation instead of treating all contact as risky, because the tension forecast distinguishes harmful overstretching from acceptable contact.
The planner's tension constraint directly reduces the risk of damaging the DLO or the robot, which matters in safety-critical applications such as assembly and surgical suturing.
Allowing necessary contact while still enforcing safety can raise task success compared with conservative contact-avoiding planners.
The same optimization structure can be reused for a variety of robot morphologies and trajectory families, with only the learned prediction model retrained for a new DLO task.

Reading between the lines

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

A formal safety certificate would require replacing the finite-test-set $\varepsilon$ with a guaranteed uniform error bound, for example via conformal prediction or verified neural-network bounds; the paper's implementation is best read as an empirical safety margin rather than a mathematical certificate.
The same joint shape-and-tension forecast could be used as an online monitor: a wrist force or torque sensor could check whether true tension tracks the predicted profile and trigger a re-plan if the discrepancy approaches $\varepsilon$, helping close the sim-to-real gap.
Model accuracy and safety margin trade directly against task success: a larger $\varepsilon$ shrinks the feasible trajectory set, so improving tension prediction through better contact modeling or more data should raise the success rate in tight environments.

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.

Desk Editor's Note

read the letter

Colleague,

The headline is that the paper's integration is plausible but the certification claim is not supported. The safety constraint Eq. (12) is conditioned on Assumption 3's epsilon, and the implementation sets epsilon to the maximum residual on the test set. That is a finite-sample statistic, not an upper bound over the planner's continuous trajectory family or over the randomized task distribution. Without a genuine bound (Lipschitz, conformal, verified), the zero-overextension results in Table I are an empirical consequence of the constraint being enforced with a chosen margin, not a certificate.

What is new and worth credit: jointly predicting DLO shape and tension with an LSTM and feeding both into an ARMOUR-style polynomial zonotope optimizer is a sensible combination that I haven't seen in the two cited baselines. The contact-aware projection and the cost that includes DLO state are reasonable design choices. The simulation results are internally consistent and show the tension constraint reduces overextensions relative to ARMOUR, and the method is less conservative than Learning Where to Trust. That is a real, if incremental, empirical contribution.

Where it goes soft: the main flaw is above, and it is load-bearing. The paper would be fine if reframed as a heuristic safety-constrained planner with uncertainty quantification, but the title and abstract promise certification. Additional soft spots: the contact projection has no provable effect on tension error; the evaluation is simulation-only with no code or data; and the reference list includes some unrelated self-citations (the drying papers), which does not help credibility.

Bottom line: this deserves serious peer review, but with a clear directive to the authors to either supply a real certificate or strip the certification language. The empirical framework is salvageable and the problem is important. I would not cite it as a certified method, but I might cite it as an example of learning-based tension-aware planning if the framing is fixed.

Bring it to reading group? Maybe, as a case study in why test-set error is not a safety certificate.

Referee Report

3 major / 5 minor

Summary. The paper proposes a motion planning and control framework for deformable linear object (DLO) manipulation that jointly predicts DLO shape and tension with an LSTM, then integrates these predictions into a real-time trajectory optimizer based on polynomial zonotopes and the ARMOUR reachability framework. The optimizer enforces joint, collision, and tension safety constraints; tension safety is handled through a constraint of the form ||f_pred|| + ε ≤ f_lim, where ε is called a certified error bound. The method is evaluated in a PyBullet wire-harness assembly task over 100 randomized trials, reporting 76 task successes, zero robot collisions, and zero DLO overextensions, compared to 54 successes/26 overextensions for ARMOUR and 38 successes/3 overextensions for Learning Where to Trust.

Significance. If the safety certification were valid, this would be an important bridge between learned deformable-object models and reachability-based planning. The empirical results are promising and the combination of learned tension prediction with an ARMOUR-style planner is a sensible heuristic. However, the paper provides no real verification or code, and the central term "certifiable" is not supported by the evidence presented; the zero-violation result in Table I is to a large extent a consequence of enforcing the proposed constraint rather than independent evidence of safety.

major comments (3)

[II.C.4 (Assumption 3 and Eq. (12))] The central safety claim rests entirely on Assumption 3, which postulates a certified error bound ε satisfying ||f_true(t_i;k) - f(t_i;k)|| ≤ ε for all time steps and all trajectory parameters k. The implementation then sets ε to the maximum residual observed on the test set. A finite test-set maximum is not an upper bound over the compact trajectory family K, the randomized engine poses, or the closed-loop receding-horizon state distribution encountered at execution; without a Lipschitz bound, a conformal guarantee, or a reachability-based bound, Eq. (12) does not imply ||f_true|| ≤ f_lim. The paper therefore does not justify the word "certifiably" in the title, and Table I's zero-overextension outcome is conditional on a particular choice of ε rather than a certificate.
[Table I and Section III.C] The zero-overextension result is a direct consequence of constraining the planner with Eq. (12), so it cannot be used as evidence that the safety guarantee holds. To substantiate the certification claim, the authors should report the distribution of residual magnitudes on held-out and out-of-distribution task instances, the worst-case residual over the planner's trajectory family, the observed gap between ||f_hat|| + ε and f_lim, and ideally show a comparison against a baseline with a genuinely certified bound (e.g., conformal or Lipschitz-based). Without such evidence, the result is an empirical observation, not a verification of Assumption 3.
[II.A.2 (Contact-aware modeling)] The contact handler projects penetrated nodes onto obstacle surfaces using a QP, but the manuscript does not specify whether tension is predicted before or after this projection. If tension is predicted from the original LSTM output, the projected shape changes are not accounted for in the error model; if tension is predicted after projection, the projection step itself becomes part of the learned pipeline and must be covered by the error bound. Either way, the missing specification leaves Assumption 3 without support for the contact-rich scenarios that are the focus of the paper.

minor comments (5)

[Section II.C.4] The notation N_t is used in Eq. (6) as if it were defined, but it is only later used in Eqs. (8)-(12); please define it explicitly as the set of time indices.
[Section II.C.3] The variables n_p and n_t are both introduced in the cost function (Eq. 5) but their relationship is unclear; the sum runs to n_p while the time horizon is divided into n_t subintervals, which should be stated more clearly.
[Table I] The column heading "Failed w/o Violation" should be written out as "Failed without Violation" for clarity.
[Section III.A.3] Please specify how the wire harness tension is extracted from PyBullet (e.g., constraint force magnitude, elastic force at nodes), since the safety constraint depends on this value.
[Section II.C.4] The norm used in Eq. (6) and Assumption 3 is not specified; because the constraint is on ||f||, define whether this is the Euclidean norm or a componentwise maximum, as the error bound may follow a different distribution for each.

Circularity Check

1 steps flagged · score 6.0 of 10

The DLO safety guarantee reduces to the fitted test-set statistic ε, so the reported zero overextension is enforced by construction rather than independently predicted.

fitted input called prediction [Section II.C.4, Assumption 3 and Eq. (12); Section III.C, Table I]
"Under this assumption, we impose the following certifiably safe constraint: || f(t_i;k) || + ε ≤ f_lim, ∀i ∈ N_t. This ensures that even in the worst case, where the model underestimates the true tension by up to ε, the actual tension will not exceed the safety limit. In our implementation, ε is estimated from the maximum residual observed on the test set."

The safety guarantee's load-bearing content is the 'certified error bound' ε, but ε is not certified: a maximum residual on a finite test set is not a uniform bound over the compact trajectory family K, randomized task configurations, or the closed-loop execution distribution. The reported 'DLO Overextension: 0' in Table I is then a logical consequence of constraint (12) plus feasibility: every executed trajectory is required to satisfy ||f_hat|| + ε ≤ f_lim, and with the assumed bound this entails f_true ≤ f_lim. The zero-overextension result is therefore built into the optimization by the fitted input ε, rather than being an independent prediction that validates the certificate.

full rationale

The paper's formal step is a valid conditional implication: if Assumption 3 holds with a certified ε, then constraint (12) guarantees f_true ≤ f_lim. That implication is not circular in itself. The circularity arises when the word 'certified' is attached to an ε that is merely 'estimated from the maximum residual observed on the test set.' The same fitted statistic then serves as the certificate in planning, and the zero-overextension row in Table I is presented as evidence of safety, even though it is entailed by the constraint for any feasible executed trajectory whenever the model error happens to remain within ε. This is a fitted input renamed as a certificate: the central safety claim reduces to the unverified bound ε. There is no load-bearing self-citation: the ARMOUR reachability machinery provides independent support for the robot-side guarantees, and the remaining DLO-specific contribution is the tension prediction plus the ε-constrained optimization. The comparison against baselines is meaningful for task success and for showing that adding the tension constraint reduces overextensions, but it does not establish the claimed certificate over the planner's continuous trajectory family. Overall, the central safety claim is partially circular: it depends by construction on the fitted ε, while the independent components (LSTM tension prediction, reachability-based trajectory optimization) remain genuine but do not by themselves supply the stated safety certificate.

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

The safety certificate rests entirely on Assumption 3, which is unverified. The learned model and the margin epsilon are fitted to simulation data, so the central claim is conditional. The remaining axioms are standard background assumptions for the robot and the DLO interaction model.

free parameters (2)

epsilon (tension error bound) = maximum residual on the test set
Constraint (12) uses this margin to certify safety; because it is a finite-sample statistic rather than a proven bound, the certificate depends on a fitted value.
LSTM model weights = learned from 1,500 training trajectories
The predicted shape and tension entering the cost function and safety constraint are outputs of a learned network; no architecture, checkpoints, or code are provided.

assumptions (4)

domain assumption Assumption 1: the robot has revolute joints, encoders for joint positions and velocities, and is fully actuated.
Defines the robot model used in the manipulator dynamics (3).
domain assumption Assumption 2: the DLO's reaction force on the robot is an external force at the end-effector equal in magnitude to the DLO tension.
Simplifies DLO-robot interaction to a point force; neglects distributed contacts and bending moments.
ad hoc to paper Assumption 3: there exists a certified error bound epsilon such that || f_true(t_i;k) - f(t_i;k) || <= epsilon for all t_i and all trajectory parameters k.
This is the load-bearing safety certificate; the paper only estimates epsilon as the maximum test-set residual, so the guarantee is assumed rather than established.
standard math Standard rigid-body manipulator equations (3) with inertial parameter set Delta.
Borrowed from a standard robotics textbook and used by the ARMOUR reachability analysis [25].

how reviews work

0 comments

Cite this review

Pith. "Pith review of Certifiably Safe Manipulation of Deformable Linear Objects via Joint Shape and Tension Prediction." pith.science (2026). https://pith.science/paper/LWJY7CFI

@misc{pith2026250513889,
  author       = {Pith},
  title        = {Pith review of: Certifiably Safe Manipulation of Deformable Linear Objects via Joint Shape and Tension Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LWJY7CFI}},
  note         = {Machine review of arXiv:2505.13889}
}

read the original abstract

Manipulating deformable linear objects (DLOs) is challenging due to their complex dynamics and the need for safe interaction in contact-rich environments. Most existing models focus on shape prediction alone and fail to account for contact and tension constraints, which can lead to damage to both the DLO and the robot. In this work, we propose a certifiably safe motion planning and control framework for DLO manipulation. At the core of our method is a predictive model that jointly estimates the DLO's future shape and tension. These predictions are integrated into a real-time trajectory optimizer based on polynomial zonotopes, allowing us to enforce safety constraints throughout the execution. We evaluate our framework on a simulated wire harness assembly task using a 7-DOF robotic arm. Compared to state-of-the-art methods, our approach achieves a higher task success rate while avoiding all safety violations. The results demonstrate that our method enables robust and safe DLO manipulation in contact-rich environments.

Figures

Figures reproduced from arXiv: 2505.13889 by the authors.

**Figure 2.** Overview of the proposed framework. Starting from an initial state, the robot generates a set of parameterized trajectories and uses a learned predictive [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 3.** a) is an overview of the discretized configuration of DLO. Blue circles [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Architecture of the DLO shape and tension prediction model. [PITH_FULL_IMAGE:figures/full_fig_p003_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

XPG-RL: Reinforcement Learning with Explainable Priority Guidance for Efficiency-Boosted Mechanical Search
cs.RO 2025-04 conditional novelty 5.0 of 10

XPG-RL learns adaptive thresholds for switching between grasping, occlusion removal, and viewpoint adjustment, improving mechanical search efficiency by up to 4.5x over baselines.

Reference graph

Works this paper leans on

27 extracted references · 22 canonical work pages · cited by 1 Pith paper

[1]

Dynamic modeling and control of deformable linear objects for single-arm and dual-arm robot manipulations,

N. Lv, J. Liu, and Y . Jia, “Dynamic modeling and control of deformable linear objects for single-arm and dual-arm robot manipulations,”IEEE Transactions on Robotics, vol. 38, no. 4, pp. 2341–2353, 2022

work page 2022
[2]

A practical solution to deformable linear object manipulation: A case study on cable harness connection,

H. Zhou, S. Li, Q. Lu, and J. Qian, “A practical solution to deformable linear object manipulation: A case study on cable harness connection,” in2020 5th International Conference on Advanced Robotics and Mecha- tronics (ICARM). IEEE, 2020, pp. 329–333

work page 2020
[3]

A novel robotic suturing system for flexible endoscopic surgery,

L. Cao, X. Li, P. T. Phan, A. M. H. Tiong, J. Liu, and S. J. Phee, “A novel robotic suturing system for flexible endoscopic surgery,” in2019 International Conference on Robotics and Automation (ICRA). IEEE, 2019, pp. 1514–1520

work page 2019
[4]

Modeling, learning, perception, and control methods for deformable object manipulation,

H. Yin, A. Varava, and D. Kragic, “Modeling, learning, perception, and control methods for deformable object manipulation,”Science Robotics, vol. 6, no. 54, p. eabd8803, 2021

2021
[5]

Robotic manipulation planning for shaping deformable linear objects withenvironmental contacts,

J. Zhu, B. Navarro, R. Passama, P. Fraisse, A. Crosnier, and A. Cheru- bini, “Robotic manipulation planning for shaping deformable linear objects withenvironmental contacts,”IEEE Robotics and Automation Letters, vol. 5, no. 1, pp. 16–23, 2019

work page 2019
[6]

Four-arm collaboration: Two dual- arm robots work together to manipulate tethered tools,

D. Sanchez, W. Wan, and K. Harada, “Four-arm collaboration: Two dual- arm robots work together to manipulate tethered tools,”IEEE/ASME Transactions on Mechatronics, vol. 27, no. 5, pp. 3286–3296, 2021

work page 2021
[7]

Identification of spring parame- ters for deformable object simulation,

B. Lloyd, G. Sz ´ekely, and M. Harders, “Identification of spring parame- ters for deformable object simulation,”IEEE transactions on visualiza- tion and computer graphics, vol. 13, no. 5, pp. 1081–1094, 2007

work page 2007
[8]

Robotic manipulation of deformable rope-like objects using differentiable compliant position- based dynamics,

F. Liu, E. Su, J. Lu, M. Li, and M. C. Yip, “Robotic manipulation of deformable rope-like objects using differentiable compliant position- based dynamics,”IEEE Robotics and Automation Letters, vol. 8, no. 7, pp. 3964–3971, 2023

2023

Show all 27 references

[9]

Vega: non-linear fem deformable object simulator,

F. S. Sin, D. Schroeder, and J. Barbi ˇc, “Vega: non-linear fem deformable object simulator,” inComputer Graphics Forum, vol. 32, no. 1. Wiley Online Library, 2013, pp. 36–48

2013
[10]

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,”IEEE robotics and automation letters, vol. 5, no. 2, pp. 2372–2379, 2020

2020
[11]

Learning to propagate interaction effects for modeling deformable linear objects dynamics,

Y . Yang, J. A. Stork, and T. Stoyanov, “Learning to propagate interaction effects for modeling deformable linear objects dynamics,” in2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2021, pp. 1950–1957

2021
[12]

Offline-online learning of deformation model for cable manipulation with graph neural networks,

C. Wang, Y . Zhang, X. Zhang, Z. Wu, X. Zhu, S. Jin, T. Tang, and M. Tomizuka, “Offline-online learning of deformation model for cable manipulation with graph neural networks,”IEEE Robotics and Automation Letters, vol. 7, no. 2, pp. 5544–5551, 2022

2022
[13]

Differentiable discrete elastic rods for real-time modeling of deformable linear objects,

Y . Chen, Y . Zhang, Z. Brei, T. Zhang, Y . Chen, J. Wu, and R. Vasudevan, “Differentiable discrete elastic rods for real-time modeling of deformable linear objects,”arXiv preprint arXiv:2406.05931, 2024

2024 arXiv
[14]

Xpg-rl: Reinforcement learning with explainable priority guidance for efficiency-boosted mechanical search,

Y . Zhang, S. Li, and E. Shrestha, “Xpg-rl: Reinforcement learning with explainable priority guidance for efficiency-boosted mechanical search,” arXiv preprint arXiv:2504.20969, 2025

2025 arXiv
[15]

Feel the tension: Manipulation of deformable linear objects in environments with fixtures using force information,

F. S ¨uberkr¨ub, R. Laezza, and Y . Karayiannidis, “Feel the tension: Manipulation of deformable linear objects in environments with fixtures using force information,” in2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2022, pp. 11 216–11 222

2022
[16]

Edo-net: Learning elastic properties of deformable objects from graph dynamics,

A. Longhini, M. Moletta, A. Reichlin, M. C. Welle, D. Held, Z. Erickson, and D. Kragic, “Edo-net: Learning elastic properties of deformable objects from graph dynamics,” in2023 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2023, pp. 3875–3881

2023
[17]

Multi-modal data fusion for moisture content prediction in apple drying,

S. Li and C. Shao, “Multi-modal data fusion for moisture content prediction in apple drying,”arXiv preprint arXiv:2504.07465, 2025

2025
[18]

Multi-modal fusion of in-situ video data and process parame- ters for online forecasting of cookie drying readiness,

——, “Multi-modal fusion of in-situ video data and process parame- ters for online forecasting of cookie drying readiness,”arXiv preprint arXiv:2504.15599, 2025

2025 arXiv
[19]

Motion planning for robotic manipulation of deformable linear objects,

M. Saha and P. Isto, “Motion planning for robotic manipulation of deformable linear objects,” inProceedings 2006 IEEE International Conference on Robotics and Automation, 2006. ICRA 2006.IEEE, 2006, pp. 2478–2484

2006
[20]

An obstacle-based rapidly-exploring random tree,

S. Rodriguez, X. Tang, J.-M. Lien, and N. M. Amato, “An obstacle-based rapidly-exploring random tree,” inProceedings 2006 IEEE International Conference on Robotics and Automation, 2006. ICRA 2006.IEEE, 2006, pp. 895–900

2006
[21]

Manipulating deformable objects by interleaving prediction, planning, and control,

D. McConachie, A. Dobson, M. Ruan, and D. Berenson, “Manipulating deformable objects by interleaving prediction, planning, and control,” The International Journal of Robotics Research, vol. 39, no. 8, pp. 957– 982, 2020

2020
[22]

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,”Science Robotics, vol. 6, no. 54, p. eabd8170, 2021

2021
[23]

Softgym: Benchmarking deep reinforcement learning for deformable object manipulation,

X. Lin, Y . Wang, J. Olkin, and D. Held, “Softgym: Benchmarking deep reinforcement learning for deformable object manipulation,” in Conference on Robot Learning. PMLR, 2021, pp. 432–448

2021
[24]

M. W. Spong, S. Hutchinson, M. Vidyasagaret al.,Robot modeling and control. Wiley New York, 2006, vol. 3

2006
[25]

Can’t touch this: Real-time, safe motion planning and control for manipulators under uncertainty,

J. Michaux, P. Holmes, B. Zhang, C. Chen, B. Wang, S. Sahgal, T. Zhang, S. Dey, S. Kousik, and R. Vasudevan, “Can’t touch this: Real-time, safe motion planning and control for manipulators under uncertainty,”arXiv preprint arXiv:2301.13308, 2023

2023 arXiv
[26]

Safe, aggressive quadrotor flight via reachability-based trajectory design,

S. Kousik, P. Holmes, and R. Vasudevan, “Safe, aggressive quadrotor flight via reachability-based trajectory design,” inDynamic Systems and Control Conference, vol. 59162. American Society of Mechanical Engineers, 2019, p. V003T19A010

2019
[27]

Refine: Reachability-based trajectory design using robust feedback linearization and zonotopes,

J. Liu, Y . S. Shao, L. Lymburner, H. Qin, V . Kaushik, L. Trang, R. Wang, V . Ivanovic, H. E. Tseng, and R. Vasudevan, “Refine: Reachability-based trajectory design using robust feedback linearization and zonotopes,” IEEE Transactions on Robotics, vol. 40, pp. 2060–2080, 2024. 5

2024

Pith tools

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

[1] [1]

Dynamic modeling and control of deformable linear objects for single-arm and dual-arm robot manipulations,

N. Lv, J. Liu, and Y . Jia, “Dynamic modeling and control of deformable linear objects for single-arm and dual-arm robot manipulations,”IEEE Transactions on Robotics, vol. 38, no. 4, pp. 2341–2353, 2022

work page 2022

[2] [2]

A practical solution to deformable linear object manipulation: A case study on cable harness connection,

H. Zhou, S. Li, Q. Lu, and J. Qian, “A practical solution to deformable linear object manipulation: A case study on cable harness connection,” in2020 5th International Conference on Advanced Robotics and Mecha- tronics (ICARM). IEEE, 2020, pp. 329–333

work page 2020

[3] [3]

A novel robotic suturing system for flexible endoscopic surgery,

L. Cao, X. Li, P. T. Phan, A. M. H. Tiong, J. Liu, and S. J. Phee, “A novel robotic suturing system for flexible endoscopic surgery,” in2019 International Conference on Robotics and Automation (ICRA). IEEE, 2019, pp. 1514–1520

work page 2019

[4] [4]

Modeling, learning, perception, and control methods for deformable object manipulation,

H. Yin, A. Varava, and D. Kragic, “Modeling, learning, perception, and control methods for deformable object manipulation,”Science Robotics, vol. 6, no. 54, p. eabd8803, 2021

2021

[5] [5]

Robotic manipulation planning for shaping deformable linear objects withenvironmental contacts,

J. Zhu, B. Navarro, R. Passama, P. Fraisse, A. Crosnier, and A. Cheru- bini, “Robotic manipulation planning for shaping deformable linear objects withenvironmental contacts,”IEEE Robotics and Automation Letters, vol. 5, no. 1, pp. 16–23, 2019

work page 2019

[6] [6]

Four-arm collaboration: Two dual- arm robots work together to manipulate tethered tools,

D. Sanchez, W. Wan, and K. Harada, “Four-arm collaboration: Two dual- arm robots work together to manipulate tethered tools,”IEEE/ASME Transactions on Mechatronics, vol. 27, no. 5, pp. 3286–3296, 2021

work page 2021

[7] [7]

Identification of spring parame- ters for deformable object simulation,

B. Lloyd, G. Sz ´ekely, and M. Harders, “Identification of spring parame- ters for deformable object simulation,”IEEE transactions on visualiza- tion and computer graphics, vol. 13, no. 5, pp. 1081–1094, 2007

work page 2007

[8] [8]

Robotic manipulation of deformable rope-like objects using differentiable compliant position- based dynamics,

F. Liu, E. Su, J. Lu, M. Li, and M. C. Yip, “Robotic manipulation of deformable rope-like objects using differentiable compliant position- based dynamics,”IEEE Robotics and Automation Letters, vol. 8, no. 7, pp. 3964–3971, 2023

2023