pith. sign in

arxiv: 1907.09656 · v1 · pith:L7774KD6new · submitted 2019-07-23 · 💻 cs.RO · cs.CV

Grasping Using Tactile Sensing and Deep Calibration

Masoud Baghbahari , Aman Behal This is my paper

Pith reviewed 2026-05-24 18:02 UTC · model grok-4.3

classification 💻 cs.RO cs.CV
keywords tactile sensingrobot graspingforce-torque sensorsdeep calibrationfeedback controlsensor biasmanipulation
0
0 comments X

The pith

Tactile sensing with deep calibration allows robots to grasp objects effectively after an initial visual approach.

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

The paper proposes a feedback method for robot grasping that switches to force-torque tactile sensing once contact begins, using a deep learning method called Deep Calibration to correct sensor bias. This approach draws from how humans rely on touch after reaching an object, avoiding the heavy computation of continuous vision. If successful, it shows that tactile data alone can generate the actions needed to complete grasps on a real robot. The demonstration on hardware supports replacing vision with tactile feedback for the fine phase of manipulation.

Core claim

The proposed feedback approach using force-torque tactile sensing combined with Deep Calibration eliminates sensor bias and addresses the robot grasping task effectively, as demonstrated on a real robot.

What carries the argument

Deep Calibration, a deep learning framework that removes bias from force-torque tactile sensor data within a feedback loop for grasping.

If this is right

  • Grasping can proceed using only tactile sensing after initial contact without overwhelming visual computation.
  • Sensor bias in tactile readings is eliminated, leading to more reliable action generation.
  • The method works on physical robot hardware for real grasping tasks.
  • Visual perception is limited to gross reaching, with tactile taking over for interaction.

Where Pith is reading between the lines

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

  • This setup could extend to other contact-rich tasks like assembly where vision is less reliable.
  • Reducing vision use might lower the computational and energy demands of robotic systems.
  • Further testing could check performance on varied object shapes and surfaces.

Load-bearing premise

That tactile sensing provides sufficient information to generate suitable grasping actions once contact has occurred, without needing ongoing visual input.

What would settle it

Experiments showing that the robot cannot complete grasps reliably when relying solely on the calibrated tactile feedback after contact is made.

Figures

Figures reproduced from arXiv: 1907.09656 by Aman Behal, Masoud Baghbahari.

Figure 1
Figure 1. Figure 1: Robot hand interaction with object and force/torque [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: A typical vector vec coordinate transformation by rotation matrix Re b . The magnitude of vector is same in both coordinates. Assuming a Jacobian matrix with number of rows equal or greater than 6, the interaction force-torque F on end-effector can be retrieved by the Moore-Penrose inverse of the Jacobian matrix: F = (JJT ) −1Jτint (8) Since the Jacobian is a joint dependent matrix, the inverse term in som… view at source ↗
Figure 1
Figure 1. Figure 1: Using both (8) and (9), it is possible to retrieve the s [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Displacement model springiness term during interac [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 6
Figure 6. Figure 6: Bias data and fitted model of third joint [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 5
Figure 5. Figure 5: Bias data and fitted model of second joint [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 8
Figure 8. Figure 8: Bias data and fitted model of fifth joint [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗
Figure 10
Figure 10. Figure 10: Force tactile sensation profile during grasping the [PITH_FULL_IMAGE:figures/full_fig_p007_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Torque tactile sensation profile during grasping th [PITH_FULL_IMAGE:figures/full_fig_p007_11.png] view at source ↗
read the original abstract

Tactile perception is an essential ability of intelligent robots in interaction with their surrounding environments. This perception as an intermediate level acts between sensation and action and has to be defined properly to generate suitable action in response to sensed data. In this paper, we propose a feedback approach to address robot grasping task using force-torque tactile sensing. While visual perception is an essential part for gross reaching, constant utilization of this sensing modality can negatively affect the grasping process with overwhelming computation. In such case, human being utilizes tactile sensing to interact with objects. Inspired by, the proposed approach is presented and evaluated on a real robot to demonstrate the effectiveness of the suggested framework. Moreover, we utilize a deep learning framework called Deep Calibration in order to eliminate the effect of bias in the collected data from the robot sensors.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes a feedback approach for robot grasping that switches from visual perception (for gross reaching) to force-torque tactile sensing once contact occurs. It introduces a deep learning method termed Deep Calibration to remove sensor bias from the tactile data and reports evaluation on a physical robot to show the framework's effectiveness.

Significance. A working tactile-only post-contact controller with bias removal could reduce computational load compared with continuous vision, but the absence of any quantitative results, baselines, success rates, or implementation details in the provided text makes it impossible to judge whether the contribution is incremental or substantial.

major comments (2)
  1. [Abstract] Abstract: the claim that the approach 'addresses the robot grasping task effectively' and that Deep Calibration 'eliminates the effect of bias' is asserted without any supporting metrics, ablation studies, or comparison to uncalibrated tactile control; this is load-bearing for the central effectiveness claim.
  2. [Abstract] Abstract: no description is given of the feedback law, network architecture, training procedure, or loss function for Deep Calibration, so it is impossible to determine whether the bias removal is a learned mapping or reduces to a simple offset; this prevents verification of the method's soundness.
minor comments (1)
  1. [Abstract] The sentence 'Inspired by, the proposed approach...' is grammatically incomplete and should be revised for clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed comments on the abstract. We agree that the abstract can be strengthened for clarity and will revise it accordingly while preserving the high-level nature of the summary. The full manuscript contains the evaluation on the physical robot and technical descriptions of the method.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the approach 'addresses the robot grasping task effectively' and that Deep Calibration 'eliminates the effect of bias' is asserted without any supporting metrics, ablation studies, or comparison to uncalibrated tactile control; this is load-bearing for the central effectiveness claim.

    Authors: The abstract is a concise overview; the manuscript body reports real-robot evaluation demonstrating the framework. To directly address the concern about unsupported claims, we will revise the abstract to include a short reference to the experimental outcomes (e.g., effective grasping with calibrated tactile feedback). revision: yes

  2. Referee: [Abstract] Abstract: no description is given of the feedback law, network architecture, training procedure, or loss function for Deep Calibration, so it is impossible to determine whether the bias removal is a learned mapping or reduces to a simple offset; this prevents verification of the method's soundness.

    Authors: Space constraints limit abstracts to high-level statements; the feedback law, Deep Calibration architecture, training procedure, and loss function are detailed in the main text. The approach uses a learned neural-network mapping rather than a simple offset. We will add one sentence to the abstract noting the neural-network calibration to improve verifiability. revision: partial

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper presents an engineering framework for tactile-based grasping that switches from vision after contact and applies a deep learning method named Deep Calibration for sensor bias removal. No derivation chain, equations, or first-principles predictions appear in the provided text; the approach is described as inspired by human behavior and demonstrated via real-robot evaluation without any fitted parameter being relabeled as an independent prediction or any load-bearing step reducing to a self-citation or self-definition. The central claim therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only; no free parameters, axioms, or invented entities can be identified from the provided text.

pith-pipeline@v0.9.0 · 5659 in / 947 out tokens · 39526 ms · 2026-05-24T18:02:59.812795+00:00 · methodology

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

24 extracted references · 24 canonical work pages · 3 internal anchors

  1. [1]

    Making Sense of Vision and Touch: Self-Supervised Learning of Multimodal Representations for Contact-Rich Tasks

    Michelle A Lee, Y uke Zhu, Krishnan Srinivasan, Parth Sha h, Silvio Savarese, Li Fei-Fei, Animesh Garg, and Jeannette Bohg. Mak ing sense of vision and touch: Self-supervised learning of mult imodal representations for contact-rich tasks. arXiv preprint arXiv:1810.10191 , 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  2. [2]

    Learning to Grasp Without Seeing

    Adithyavairavan Murali, Yin Li, Dhiraj Gandhi, and Abhi nav Gupta. Learning to grasp without seeing. arXiv preprint arXiv:1805.04201 , 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  3. [3]

    Human-inspired robotic grasp co ntrol with tactile sensing

    Joseph M Romano, Kaijen Hsiao, G¨ unter Niemeyer, Sachin Chitta, and Katherine J Kuchenbecker. Human-inspired robotic grasp co ntrol with tactile sensing. IEEE Transactions on Robotics , 27(6):1067–1079, 2011

    work page 2011

  4. [4]

    Robot grasp planning based on demonstr ated grasp strategies

    Y un Lin and Y u Sun. Robot grasp planning based on demonstr ated grasp strategies. The International Journal of Robotics Research , 34(1):26–42, 2015

    work page 2015

  5. [5]

    Tactile sensing for mobile manipulation

    Sachin Chitta, J¨ urgen Sturm, Matthew Piccoli, and Wolf ram Burgard. Tactile sensing for mobile manipulation. IEEE Transactions on Robotics, 27(3):558–568, 2011

    work page 2011

  6. [6]

    Reliable object handover through tactile force sensing and effort control in the shadow robot hand

    A G´ omez Egu´ ıluz, I Rano, Sonya A Coleman, and T Martin Mc Ginnity. Reliable object handover through tactile force sensing and effort control in the shadow robot hand. In 2017 IEEE International Conference on Robotics and Automation (ICRA) , pages 372–377. IEEE, 2017

    work page 2017

  7. [7]

    Control of linear and rotational slippa ge based on six-axis force/tactile sensor

    Andrea Cirillo, Pasquale Cirillo, Giuseppe De Maria, Ci ro Natale, and Salvatore Pirozzi. Control of linear and rotational slippa ge based on six-axis force/tactile sensor. In 2017 IEEE International Conference on Robotics and Automation (ICRA) , pages 1587–1594. IEEE, 2017

    work page 2017

  8. [8]

    A hybrid deep architecture for robotic grasp detection

    Di Guo, Fuchun Sun, Huaping Liu, Tao Kong, Bin Fang, and Ni ng Xi. A hybrid deep architecture for robotic grasp detection. In 2017 IEEE International Conference on Robotics and Automation (ICRA ), pages 1609–1614. IEEE, 2017

    work page 2017

  9. [9]

    Deep learning for tactile understanding from visu al and haptic data

    Y ang Gao, Lisa Anne Hendricks, Katherine J Kuchenbecker , and Trevor Darrell. Deep learning for tactile understanding from visu al and haptic data. In 2016 IEEE International Conference on Robotics and Automation (ICRA) , pages 536–543. IEEE, 2016

    work page 2016

  10. [10]

    Stable reinforcement learning with autoenc oders for tactile and visual data

    Herke van Hoof, Nutan Chen, Maximilian Karl, Patrick va n der Smagt, and Jan Peters. Stable reinforcement learning with autoenc oders for tactile and visual data. In 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages 3928–3934. IEEE, 2016

    work page 2016

  11. [11]

    Deep spatial autoencoders for visuomoto r learning

    Chelsea Finn, Xin Y u Tan, Y an Duan, Trevor Darrell, Serg ey Levine, and Pieter Abbeel. Deep spatial autoencoders for visuomoto r learning. In 2016 IEEE International Conference on Robotics and Automat ion (ICRA), pages 512–519. IEEE, 2016

    work page 2016

  12. [12]

    Learn ing robot tactile sensing for object manipulation

    Y evgen Chebotar, Oliver Kroemer, and Jan Peters. Learn ing robot tactile sensing for object manipulation. In 2014 IEEE/RSJ International Conference on Intelligent Robots and Systems , pages 3368–3375. IEEE, 2014

    work page 2014

  13. [13]

    Self-supervised regrasping using spatio-t emporal tactile features and reinforcement learning

    Y evgen Chebotar, Karol Hausman, Zhe Su, Gaurav S Sukhat me, and Stefan Schaal. Self-supervised regrasping using spatio-t emporal tactile features and reinforcement learning. In 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages 1960–1966. IEEE, 2016

    work page 2016

  14. [14]

    Le arning dynamic tactile sensing with robust -based training

    Oliver Kroemer, Christoph H Lampert, and Jan Peters. Le arning dynamic tactile sensing with robust -based training. IEEE transactions on robotics, 27(3):545–557, 2011

    work page 2011

  15. [15]

    Object identification with tactile sensors using bag-of-features

    Alexander Schneider, J¨ urgen Sturm, Cyrill Stachniss , Marco Reisert, Hans Burkhardt, and Wolfram Burgard. Object identification with tactile sensors using bag-of-features. In IROS, volume 9, pages 243–248, 2009

    work page 2009

  16. [16]

    Touch based perception for object manipulation

    Anna Petrovskaya, Oussama Khatib, Sebastian Thrun, an d Andrew Y Ng. Touch based perception for object manipulation

    work page

  17. [17]

    A measurement model fo r tracking hand-object state during dexterous manipulation

    Craig Corcoran and Robert Platt. A measurement model fo r tracking hand-object state during dexterous manipulation. In 2010 IEEE Interna- tional Conference on Robotics and Automation , pages 4302–4308. IEEE, 2010

    work page 2010

  18. [18]

    Localization an d manipulation of small parts using gelsight tactile sensing

    Rui Li, Robert Platt, Wenzhen Y uan, Andreas ten Pas, Nat han Roscup, Mandayam A Srinivasan, and Edward Adelson. Localization an d manipulation of small parts using gelsight tactile sensing . In 2014 IEEE/RSJ International Conference on Intelligent Robots a nd Systems , pages 3988–3993. IEEE, 2014

    work page 2014

  19. [19]

    Assessing grasp stability based on learn ing and haptic data

    Y asemin Bekiroglu, Janne Laaksonen, Jimmy A Jørgensen , Ville Kyrki, and Danica Kragic. Assessing grasp stability based on learn ing and haptic data. IEEE Transactions on Robotics , 27(3):616, 2011

    work page 2011

  20. [20]

    Stable grasping under pose un certainty using tactile feedback

    Hao Dang and Peter K Allen. Stable grasping under pose un certainty using tactile feedback. Autonomous Robots , 36(4):309–330, 2014

    work page 2014

  21. [21]

    Force control in object manipulationa model for the study of senso rimotor control strategies

    Dennis A Nowak, Stefan Glasauer, and Joachim Hermsd¨ or fer. Force control in object manipulationa model for the study of senso rimotor control strategies. Neuroscience & Biobehavioral Reviews , 37(8):1578– 1586, 2013

    work page 2013

  22. [22]

    Intera ctive perception: Leveraging action in perception and perceptio n in action

    Jeannette Bohg, Karol Hausman, Bharath Sankaran, Oliv er Brock, Danica Kragic, Stefan Schaal, and Gaurav S Sukhatme. Intera ctive perception: Leveraging action in perception and perceptio n in action. IEEE Transactions on Robotics , 33(6):1273–1291, 2017

    work page 2017

  23. [23]

    Evaluation of estimation approaches on the quality and robustness of collision warni ng systems

    Masoud Baghbahari and Neda Hajiakhoond. Evaluation of estimation approaches on the quality and robustness of collision warni ng systems. In SoutheastCon 2018 , pages 1–7. IEEE, 2018

    work page 2018

  24. [24]

    Real-time policy generation and its application to robot grasping

    Masoud Baghbahari, Neda Hajiakhoond, and Aman Behal. R eal-time policy generation and its application to robot grasping. arXiv preprint arXiv:1808.05244, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018