Optimal Robotic Velcro Peeling with Force Feedback

Changhyun Choi; Jiacheng Yuan; Volkan Isler

REVIEW 5 major objections 3 minor 28 references

Optimal Robotic Velcro Peeling with Force Feedback

T0 review · 5 major / 3 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Force feedback alone lets a robot peel Velcro straps near-optimally.

desk verdict A worthwhile force-only peeling pipeline with a decomposed particle filter, but the advertised closed-form optimality is never derived and the abstract's success/cost claims contradict the paper's own tables. read the letter →

arxiv 2506.05812 v1 pith:GR6Y2NIK submitted 2025-06-06 cs.RO

Jiacheng Yuan , Changhyun Choi , Volkan Isler This is my paper

classification cs.RO

keywords velcropeelingdeformableobjectmanipulationforcefeedbackparticlefilterstateestimationquasi-staticmodelenergyoptimizationrobot

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper shows that a robot can peel a Velcro strap from an unknown, arbitrarily-shaped surface using only force and end-effector position feedback. The authors derive a closed-form optimal peeling strategy for the fully-observable case, minimizing a total energy cost defined by keeping the peeled part perpendicular to the attached part. For the realistic partially-observable case, they build a particle-filter state estimator that recovers the hidden surface orientation from noisy force and position readings, and a heuristic controller that alternates exploratory and exploitative actions to keep the estimate healthy. In simulation with complex geometries and sensor noise, the method achieves a 100% success rate while staying within 80% of the optimal energy cost, far ahead of reinforcement-learning baselines.

What carries the argument

The load-bearing device is the taut-peeled-part geometry: the peeled segment is treated as a straight line of length $r$ at angle $\phi$, so the five-dimensional state $X = [h_x, h_y, \theta, \phi, r]$ and the action $A = [\alpha, d, s, \delta\phi]$ satisfy the algebraic transition constraints of Eq. (3). This geometry yields the closed-form fully-observable solution, the observation model in Eqs. (5–7), and the energy cost of Eq. (9). The particle filter's state-space decomposition and the controller's health index, which triggers non-peeling exploratory actions when the weight distribution signals sample impoverishment, carry the partial-observability result.

What would settle it

Run the controller on a compliant or loosely-tensioned strap so the peeled segment visibly sags, and record whether the particle-filter hinge estimates diverge from the true peel line while the success rate drops below the reported 100%: if they do, the tautness premise is the failure point.

Watch

Extended reading notes

Core claim

Once the peeled segment of Velcro is assumed taut, the peeling configuration at any time is captured by five parameters: hinge position, surface orientation angle $\theta$, peel angle $\phi$, and peeled length $r$. The paper's central claim is that for this model the fully-observable peeling problem admits a closed-form optimal solution minimizing the energy cost $U(X) = 1 + \|\phi - \theta - \pi/2\|^2$, which serves as a theoretical performance limit. In the partially-observable case, a particle filter with state-space decomposition—updating $\phi$ under peeling actions, $\phi$ and hinge position under non-peeling actions, and $\theta$ via a line/arc fitting auxiliary estimator—combined with a heuristic controller yields 100% success with less than 80% energy increase over the optimal baseline across flat, arc, and corner geometries.

Load-bearing premise

The peeled part of the Velcro must remain taut for the entire task: if the strap goes slack, the straight-line geometric model, the transition equations, and the particle filter's measurement models no longer describe reality.

Editorial extensions

If this is right

A robot with only a wrist force sensor and proprioception can reliably peel straps from curved and cornered surfaces, removing the need for visual feedback during the peeling phase.
The closed-form fully-observable solution provides a reusable performance lower bound for any future Velcro-peeling controller.
The state-estimation design—updating only observable subsets per action type—is a template for other deformable-object manipulation tasks with mixed observable and hidden state parameters.
The method's success suggests that quasi-static modeling plus Bayesian filtering can outperform model-free reinforcement learning in this contact-rich manipulation regime.

Reading between the lines

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

The tautness-maintenance requirement could be relaxed by adding a slackness-aware state or by actively controlling tension, an extension the paper does not pursue.
The unobservability of $\theta$ in the transition model suggests the particle filter needs the auxiliary line/arc fitting step; a similar auxiliary estimator may be necessary in other tasks where a hidden parameter is Markov but unconstrained by actions.
If the 80% energy gap holds on hardware with real force-sensor noise, tactile-only peeling could be integrated into shoe-removal and garment-assist robots operating in low-light or occluded conditions.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

First, the useful part: the paper builds a clean quasi-static model of Velcro peeling under a taut-peeled-part assumption, designs a particle filter that deliberately updates only subsets of the state depending on action type, and shows in simulation that this estimator plus a heuristic controller beats two RL baselines on flat, arc, and corner shapes. That is a real, if incremental, contribution to force-only deformable-object manipulation. The state-space decomposition and the auxiliary PCA/arc fitting for the hidden hinge orientation are the most thought-through pieces. The citation pattern is fair: related work covers contact manipulation and active perception, and the self-citation to their own RL peeling paper is relevant.

Now the soft spots, in rough order of size. The central advertised result, a closed-form optimal solution for the fully observable case, is never actually presented. No theorem, equation, or derivation gives the optimal policy or its value. When you get to the evaluation, the Full-Obs baseline is described, in the authors' own words, as 'a simple heuristics' whose cost 'may act as the lower bound.' Algorithm 1 is heuristic. So the abstract's phrase 'optimal solution in closed-form which minimizes the total energy cost' is not supported by anything in the paper. This is the load-bearing claim.

Second, the headline numbers conflict with the tables. The abstract says '100% success rate with less than 80% increase in energy cost compared to the optimal solution.' Table II gives Corner success as 98.5%, not 100%. Table I gives Corner cost 103.7 versus Full-Obs 50.6, which is about a 105% increase, not less than 80%. Flat and Arc are under 80% when measured against the Full-Obs heuristic, but the blanket statement is false.

Third, there are no error bars or confidence intervals on any of the cost or success numbers across the 200 random configurations, and several parameters needed to reproduce the simulation (the c1 and c2 weights, measurement noise sigmas, action step d, health-index thresholds, forbidden-zone angles) are not listed. The paper is simulation-only; the accompanying video is mentioned but no hardware results appear in the text.

The taut-peeled-part assumption is a genuine scope condition, but it is stated clearly and the controller actively maintains it. I would not call that a flaw; it is a limitation that belongs in the abstract.

Who is this for: people working on deformable-object manipulation and active perception with force/tactile feedback. They will find the state-estimation design useful even if they do not take the optimality claim seriously. The paper deserves a serious referee, because the method is concrete and the experiments are non-trivial. But the referee should require that the claims be corrected: drop 'optimal' or actually derive it, fix the abstract numbers, and report the missing parameters and variances. As written, the central claim does not hold up.

Referee Report

5 major / 3 minor

Summary. The paper proposes a model-based approach for robotic Velcro peeling using force and position feedback. It models the peeling state under a taut-peeled-part assumption, designs a particle filter with state-space decomposition, and presents a heuristic controller that alternates peeling and non-peeling actions. The central claimed contributions are a closed-form optimal solution for the fully observable case and a partial-observability controller achieving 100% success with less than 80% energy increase over that optimum.

Significance. If the central claims were established, the paper would provide a useful theoretical baseline for a challenging deformable-object manipulation task with limited sensing. The state estimation scheme with decomposed particle filter updates is a reasonable engineering choice, and the empirical comparison against RL baselines suggests practical advantages. However, the closed-form optimality claim is never substantiated, and the headline numbers in the abstract are directly contradicted by the paper's own tables. The analytical model is a strength, but the paper does not support its strongest claims.

major comments (5)

[Abstract and Section I] The claimed 'closed-form optimal solution which minimizes the total energy cost' is never derived or stated in the manuscript. Section IV.C explicitly describes the Full-Obs baseline as 'a simple heuristics' whose cost 'may act as the lower bound', and Algorithm 1 is heuristic. No theorem, equation, or policy expression is given for an optimal solution, so the abstract's comparison 'less than 80% increase compared to the optimal solution' is not meaningful.
[Table II vs Abstract] The abstract and Section I claim a 100% success rate, but Table II reports 98.5% for the Corner case. This is a direct numerical contradiction that must be resolved.
[Table I vs Abstract] The abstract claims 'less than 80% increase in energy cost compared to the optimal solution'. Using Table I, the Corner case has 103.7/50.6 = 2.05, i.e., a 105% increase. Even if the intended claim is an average over shapes, the wording is unconditional and is violated by the published numbers.
[Section III.C, Eq. (8)-(9)] The cost function U(X) is constructed to penalize deviation from phi - theta = pi/2, and Algorithm 1's first step enforces exactly that configuration. The manuscript does not show that this potential function corresponds to the physical energy of peeling or that the proposed policy minimizes the total cost in Eq. (9). Without an independent derivation, the 'optimality' claim is partly circular.
[Section III.A and Eq. (7)] The model relies on the peeled part being always taut and on a zero-mean Gaussian random walk for theta (Eq. 7), which limits the surface curvature. The abstract states that 'the surface geometry is arbitrary and unknown', which is stronger than these model assumptions. The paper should either restrict the claim to geometries satisfying the tautness and mild-curvature conditions or provide evidence that violations do not affect the reported performance.

minor comments (3)

[Tables I and II] The reported costs and success rates are point estimates over 200 configurations; standard deviations or confidence intervals would strengthen the quantitative claims, especially given the exact percentages in the abstract.
[Figure 6] The axis labels and units in panels (c), (d), (g), and (h) are unclear; the time-step axis is not labeled in all subplots, making it difficult to interpret the evolution of theta, phi, and r.
[References] Reference [8] is a duplicate of reference [6]; the paper should use distinct references for distinct works.

Circularity Check

1 steps flagged · score 6.0 of 10

The advertised closed-form optimal baseline is not derived: it is the tautological minimizer of a cost built around the right-angle configuration, and the 100%/80% headline numbers conflict with the paper's own tables.

self definitional [Section III.C, Eqs. (8)-(9); Section IV.C Full-Obs baseline; Algorithm 1]
"As a result, we design a potential function in Eq. 8 that tracks the angle between the peeled and attached part and adds quadratic penalization to the deviation from the right angle. Then, the cost in Eq. 9, can be defined by integrating the potential along a path. ... If the state is fully observable, a simple heuristics can be designed to keep the trajectory as close to the minimum potential region as possible. The aggregated cost in this case may act as the lower bound for the partially observable case."

The full-observation 'optimal solution' is never derived as a theorem or closed-form policy; the only candidate offered is the Full-Obs baseline, a 'simple heuristics' that keeps the trajectory near the 'minimum potential region.' Since C in Eq. (9) is the integral of U in Eq. (8), and U = 1 + ||phi-theta-pi/2||^2 has its minimum exactly at phi-theta = pi/2, the 'optimal' or 'lower-bound' behavior is the pointwise minimizer of a cost the authors constructed around that configuration. Algorithm 1 then explicitly forces phi-theta = pi/2 with the non-peeling action A = (0,0,1, pi/2-(phi-theta)). Thus the claimed optimality reduces, by construction, to the definition of the objective; no independent derivation from the state/action model is supplied.

full rationale

The one genuine circular step is the full-observability optimality claim. Eq. (8) defines U with a quadratic penalty centered at phi-theta = pi/2, Eq. (9) defines total cost as the integral of U, and the full-observation baseline is explicitly a heuristic that stays in the 'minimum potential region' of that same U; Algorithm 1 enforces the right-angle configuration. The claimed closed-form optimal solution is therefore not an independent first-principles result but the tautological minimizer of the authors' own cost. I do not count the particle-filter design as circular: it follows the external framework of Gustafsson et al. [26], and the partial-observability success rates are empirical results, not constructions. Separately, the abstract's headline '100% success rate with less than 80% increase' is internally inconsistent with Tables I and II (Corner success is 98.5%; Corner cost increase is (103.7-50.6)/50.6 ~ 105%), and the paper does not present any theorem giving the closed-form optimal policy; these are correctness and support problems, not additional circularity. Overall score 6: the central optimality claim partially reduces by construction, while the estimator and empirical evaluation retain independent content.

Assumptions & free parameters 8 free parameters · 7 assumptions · 0 invented entities

The central result rests on a quasi-static taut-strap model with several hand-chosen cost and noise parameters. The cost function is designed to prefer a perpendicular peeling angle, and the controller enforces that angle, so the optimality result is substantially baked into the cost definition. No new physical entities are introduced.

free parameters (8)

c1 = not specified
Weight for peeling cost in Eq. (9); affects all reported energy values and optimality comparisons.
c2 = not specified
Weight for non-peeling cost in Eq. (9); likewise affects energy totals.
sigma_1 = not specified
Standard deviation of force direction measurement noise in Eq. (5); controls particle filter update behavior.
sigma_2 = not specified
Standard deviation of tip position measurement noise in Eq. (6); controls filter update.
sigma_3 = not specified
Standard deviation of theta random-walk noise in Eq. (7); determines how quickly the filter trusts previous theta.
action_step_d = not specified
Length of peeling action steps in Algorithm 1; affects discretization of the geometry and cost.
health_index_thresholds = 0.1 and 0.9
Hand-chosen particle weight thresholds in the health index z; determine when extra exploration actions are taken.
forbidden_zone_angles = 5 and 175 degrees
Hand-chosen failure thresholds in Section IV.B; define when an episode is marked failed.

assumptions (7)

domain assumption The peeled part of the Velcro is always taut.
Stated in Section III.A; basis for straight-line state representation and Eq. (3) geometry.
domain assumption Quasi-static dynamics: the robot moves slowly enough that inertial effects can be ignored.
Used throughout Section III; justifies force magnitude being irrelevant and purely geometric transitions.
domain assumption For small peeling steps, the local surface geometry can be approximated as a straight vector.
Invoked before Eq. (3) to derive the transition model for hinge movement.
domain assumption The force sensor is unbiased and its direction is phi + pi plus Gaussian noise.
This is Eq. (5), used to update particles from force observations.
domain assumption The tip position observation is h + r*(cos phi, sin phi) plus Gaussian noise.
This is Eq. (6), the position observation model.
ad hoc to paper The underlying surface curvature is mild enough that theta evolves as a zero-mean Gaussian random walk.
Eq. (7) imposes a radius constraint on the surface; the paper admits real surfaces like circles and corners do not fit this model and adds an auxiliary estimator.
ad hoc to paper The energy cost defined in Eq. (8)-(9) is the correct quantity to minimize for peeling.
The cost penalizes deviation from the perpendicular configuration based on human behavior; no physical energy derivation is given.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Optimal Robotic Velcro Peeling with Force Feedback." pith.science (2026). https://pith.science/paper/GR6Y2NIK

@misc{pith2026250605812,
  author       = {Pith},
  title        = {Pith review of: Optimal Robotic Velcro Peeling with Force Feedback},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GR6Y2NIK}},
  note         = {Machine review of arXiv:2506.05812}
}

read the original abstract

We study the problem of peeling a Velcro strap from a surface using a robotic manipulator. The surface geometry is arbitrary and unknown. The robot has access to only the force feedback and its end-effector position. This problem is challenging due to the partial observability of the environment and the incompleteness of the sensor feedback. To solve it, we first model the system with simple analytic state and action models based on quasi-static dynamics assumptions. We then study the fully-observable case where the state of both the Velcro and the robot are given. For this case, we obtain the optimal solution in closed-form which minimizes the total energy cost. Next, for the partially-observable case, we design a state estimator which estimates the underlying state using only force and position feedback. Then, we present a heuristics-based controller that balances exploratory and exploitative behaviors in order to peel the velcro efficiently. Finally, we evaluate our proposed method in environments with complex geometric uncertainties and sensor noises, achieving 100% success rate with less than 80% increase in energy cost compared to the optimal solution when the environment is fully-observable, outperforming the baselines by a large margin.

Figures

Figures reproduced from arXiv: 2506.05812 by the authors.

**Figure 1.** Velcro peeling diagram illustrating a robotic manipulator grasping the tip of the peeled part (green) of the Velcro strap and the rest of Velcro is applied on a rigid object with a varying contour. Hinge refers to the connecting point of the peeled and attached part with a vector pointing the orientation of the following segments. challenges require the peeling controller to introduce a tradeoff between efficiency … view at source ↗

**Figure 2.** Illustration of the Velcro peeling state. B. Active Perception Active perception [18] refers to scenarios in which a robot actively interacts with its environment to improve its perception and understanding. Rather than performing passive observation, the robot strategically plans and performs actions to gather information, thereby reducing uncertainty about the state of the environment. Active perception allows ro… view at source ↗

**Figure 4.** [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗

Figures from the paper (2 more)

**Figure 5.** Figure 5: Examples of various Velcro shapes and poses generated for evaluation with the underlying surface (black), the gripper (yellow), the peeled part (gray), and a coordinate frame (red/green) centered at the starting point of the Velcro. (a) Flat Velcro with a negative tilt…

**Figure 6.** Figure 6: Examples of how the particle filter evolves during an episode of Velcro peeling. The first column (a, e) illustrates the shape of an arc Velcro and a corner Velcro. The second column (b, f) plots the hinge trajectory of sim (red) and all the particles (blue) for only a…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

28 extracted references · 23 canonical work pages

[1]

Multi-step recurrent q-learning for robotic velcro peeling,

J. Yuan, N. H ¨ani, and V . Isler, “Multi-step recurrent q-learning for robotic velcro peeling,” in2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2021, pp. 6657–6663

work page 2021
[2]

A survey of robot manipulation in contact,

M. Suomalainen, Y . Karayiannidis, and V . Kyrki, “A survey of robot manipulation in contact,”Robotics and Autonomous Systems, vol. 156, p. 104224, 2022

2022
[3]

Active planar mass distribution estimation with robotic manipulation,

J. Yuan, C. Choi, E. B. Tadmor, and V . Isler, “Active planar mass distribution estimation with robotic manipulation,” in2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2023, pp. 10 785–10 791

work page 2023
[4]

Cognition-enabled robotic wiping: Representation, planning, execu- tion, and interpretation,

D. Leidner, G. Bartels, W. Bejjani, A. Albu-Sch ¨affer, and M. Beetz, “Cognition-enabled robotic wiping: Representation, planning, execu- tion, and interpretation,”Robotics and Autonomous Systems, vol. 114, pp. 199–216, 2019

work page 2019
[5]

Reform: A robot learning sandbox for deformable linear object manipulation,

R. Laezza, R. Gieselmann, F. T. Pokorny, and Y . Karayiannidis, “Reform: A robot learning sandbox for deformable linear object manipulation,” in2021 IEEE international conference on robotics and automation (ICRA). IEEE, 2021, pp. 4717–4723

work page 2021
[7]

Deep reinforce- ment learning with smooth policy update: Application to robotic cloth manipulation,

Y . Tsurumine, Y . Cui, E. Uchibe, and T. Matsubara, “Deep reinforce- ment learning with smooth policy update: Application to robotic cloth manipulation,”Robotics and Autonomous Systems, vol. 112, pp. 72– 83, 2019

work page 2019
[8]

Static stability of robotic fabric strip folding,

V . Petr´ık, V . Smutn`y, and V . Kyrki, “Static stability of robotic fabric strip folding,”IEEE/ASME Transactions on Mechatronics, vol. 25, no. 5, pp. 2493–2500, 2020

work page 2020
[9]

Comparative peg- in-hole testing of a force-based manipulation controlled robotic hand,

K. Van Wyk, M. Culleton, J. Falco, and K. Kelly, “Comparative peg- in-hole testing of a force-based manipulation controlled robotic hand,” IEEE Transactions on Robotics, vol. 34, no. 2, pp. 542–549, 2018

work page 2018

Show all 28 references

[10]

Peg-in-hole assembly based on two-phase scheme and f/t sensor for dual-arm robot,

X. Zhang, Y . Zheng, J. Ota, and Y . Huang, “Peg-in-hole assembly based on two-phase scheme and f/t sensor for dual-arm robot,” Sensors, vol. 17, no. 9, p. 2004, 2017

2004
[11]

Learning robot tactile sensing for object manipulation,

Y . Chebotar, O. Kroemer, and J. Peters, “Learning robot tactile sensing for object manipulation,” in2014 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, 2014, pp. 3368–3375

2014
[12]

Learning compliant manipulation through kinesthetic and tactile human-robot interaction,

K. Kronander and A. Billard, “Learning compliant manipulation through kinesthetic and tactile human-robot interaction,”IEEE trans- actions on haptics, vol. 7, no. 3, pp. 367–380, 2013

2013
[13]

Learning multiple strategies to perform a valve turning with under- water currents using an i-auv,

A. Carrera, N. Palomeras, N. Hurt ´os, P. Kormushev, and M. Carreras, “Learning multiple strategies to perform a valve turning with under- water currents using an i-auv,” inOCEANS 2015-Genova. IEEE, 2015, pp. 1–8

2015
[14]

An adaptive control approach for opening doors and drawers under uncertainties,

Y . Karayiannidis, C. Smith, F. E. V . Barrientos, P. ¨Ogren, and D. Kragic, “An adaptive control approach for opening doors and drawers under uncertainties,”IEEE Transactions on Robotics, vol. 32, no. 1, pp. 161–175, 2016

2016
[15]

Learning robot manipulation tasks with task-parameterized semitied hidden semi-markov model,

A. K. Tanwani and S. Calinon, “Learning robot manipulation tasks with task-parameterized semitied hidden semi-markov model,”IEEE Robotics and Automation Letters, vol. 1, no. 1, pp. 235–242, 2016

2016
[16]

Learning contact-rich ma- nipulation skills with guided policy search (2015),

S. Levine, N. Wagener, and P. Abbeel, “Learning contact-rich ma- nipulation skills with guided policy search (2015),”arXiv preprint arXiv:1501.05611, 2015

2015 arXiv
[17]

Reinforcement learning for improving imitated in-contact skills,

M. Hazara and V . Kyrki, “Reinforcement learning for improving imitated in-contact skills,” in2016 IEEE-RAS 16th International Conference on Humanoid Robots (Humanoids). IEEE, 2016, pp. 194–201

2016
[18]

Interactive perception: Leveraging action in perception and perception in action,

J. Bohg, K. Hausman, B. Sankaran, O. Brock, D. Kragic, S. Schaal, and G. S. Sukhatme, “Interactive perception: Leveraging action in perception and perception in action,”IEEE Transactions on Robotics, vol. 33, no. 6, pp. 1273–1291, 2017

2017
[19]

Learning haptic representation of objects,

L. Natale, G. Metta, and G. Sandini, “Learning haptic representation of objects,” inInternational conference on intelligent manipulation and grasping. Genoa, 2004, p. 43

2004
[20]

Estimation of inertial parameters of manipulator loads and links,

C. G. Atkeson, C. H. An, and J. M. Hollerbach, “Estimation of inertial parameters of manipulator loads and links,”The International Journal of Robotics Research, vol. 5, no. 3, pp. 101–119, 1986

1986
[21]

Model-free robot manipulation of doors and drawers by means of fixed-grasps,

Y . Karayiannidis, C. Smith, F. E. Vina, P. ¨Ogren, and D. Kragic, “Model-free robot manipulation of doors and drawers by means of fixed-grasps,” in2013 IEEE International Conference on Robotics and Automation. IEEE, 2013, pp. 4485–4492

2013
[22]

Pre-and post- contact policy decomposition for planar contact manipulation under uncertainty,

M. C. Koval, N. S. Pollard, and S. S. Srinivasa, “Pre-and post- contact policy decomposition for planar contact manipulation under uncertainty,”The International Journal of Robotics Research, vol. 35, no. 1-3, pp. 244–264, 2016

2016
[23]

Efficient touch based localization through submodularity,

S. Javdani, M. Klingensmith, J. A. Bagnell, N. S. Pollard, and S. S. Srinivasa, “Efficient touch based localization through submodularity,” in2013 IEEE International Conference on Robotics and Automation. IEEE, 2013, pp. 1828–1835

2013
[24]

Gaussian process implicit surfaces for shape estimation and grasping,

S. Dragiev, M. Toussaint, and M. Gienger, “Gaussian process implicit surfaces for shape estimation and grasping,” in2011 IEEE Interna- tional Conference on Robotics and Automation. IEEE, 2011, pp. 2845–2850

2011
[25]

Uncertainty aware grasping and tactile exploration,

——, “Uncertainty aware grasping and tactile exploration,” in2013 IEEE International conference on robotics and automation. IEEE, 2013, pp. 113–119

2013
[26]

Particle filters for positioning, navigation, and tracking,

F. Gustafsson, F. Gunnarsson, N. Bergman, U. Forssell, J. Jansson, R. Karlsson, and P.-J. Nordlund, “Particle filters for positioning, navigation, and tracking,”IEEE Transactions on signal processing, vol. 50, no. 2, pp. 425–437, 2002

2002
[27]

Novel approach to nonlinear/non-gaussian bayesian state estimation,

N. J. Gordon, D. J. Salmond, and A. F. Smith, “Novel approach to nonlinear/non-gaussian bayesian state estimation,” inIEE proceedings F (radar and signal processing), vol. 140, no. 2. IET, 1993, pp. 107–113

1993
[28]

Proximal policy optimization algorithms,

J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, “Proximal policy optimization algorithms,”arXiv preprint arXiv:1707.06347, 2017

2017 arXiv
[29]

Attention is all you need,

A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in neural information processing systems, vol. 30, 2017

2017

Pith tools

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

[1] [1]

Multi-step recurrent q-learning for robotic velcro peeling,

J. Yuan, N. H ¨ani, and V . Isler, “Multi-step recurrent q-learning for robotic velcro peeling,” in2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2021, pp. 6657–6663

work page 2021

[2] [2]

A survey of robot manipulation in contact,

M. Suomalainen, Y . Karayiannidis, and V . Kyrki, “A survey of robot manipulation in contact,”Robotics and Autonomous Systems, vol. 156, p. 104224, 2022

2022

[3] [3]

Active planar mass distribution estimation with robotic manipulation,

J. Yuan, C. Choi, E. B. Tadmor, and V . Isler, “Active planar mass distribution estimation with robotic manipulation,” in2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2023, pp. 10 785–10 791

work page 2023

[4] [4]

Cognition-enabled robotic wiping: Representation, planning, execu- tion, and interpretation,

D. Leidner, G. Bartels, W. Bejjani, A. Albu-Sch ¨affer, and M. Beetz, “Cognition-enabled robotic wiping: Representation, planning, execu- tion, and interpretation,”Robotics and Autonomous Systems, vol. 114, pp. 199–216, 2019

work page 2019

[5] [5]

Reform: A robot learning sandbox for deformable linear object manipulation,

R. Laezza, R. Gieselmann, F. T. Pokorny, and Y . Karayiannidis, “Reform: A robot learning sandbox for deformable linear object manipulation,” in2021 IEEE international conference on robotics and automation (ICRA). IEEE, 2021, pp. 4717–4723

work page 2021

[6] [7]

Deep reinforce- ment learning with smooth policy update: Application to robotic cloth manipulation,

Y . Tsurumine, Y . Cui, E. Uchibe, and T. Matsubara, “Deep reinforce- ment learning with smooth policy update: Application to robotic cloth manipulation,”Robotics and Autonomous Systems, vol. 112, pp. 72– 83, 2019

work page 2019

[7] [8]

Static stability of robotic fabric strip folding,

V . Petr´ık, V . Smutn`y, and V . Kyrki, “Static stability of robotic fabric strip folding,”IEEE/ASME Transactions on Mechatronics, vol. 25, no. 5, pp. 2493–2500, 2020

work page 2020

[8] [9]

Comparative peg- in-hole testing of a force-based manipulation controlled robotic hand,

K. Van Wyk, M. Culleton, J. Falco, and K. Kelly, “Comparative peg- in-hole testing of a force-based manipulation controlled robotic hand,” IEEE Transactions on Robotics, vol. 34, no. 2, pp. 542–549, 2018

work page 2018