pith. sign in

arxiv: 2605.04610 · v1 · submitted 2026-05-06 · 💻 cs.RO

Active Contact Sensing for Robust Robot-to-Human Object Handover

Linfeng Li , Lin Shao , David Hsu This is my paper

Pith reviewed 2026-05-08 16:07 UTC · model grok-4.3

classification 💻 cs.RO
keywords robot handoveractive contact sensingfirm grasp detectionBayesian linear modelforce sensinghuman-robot interactionpiecewise linear mappingobject transfer
0
0 comments X

The pith

Robots detect firm human grasps during object handover by applying small motions and modeling the resulting forces.

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

The paper shows that passive sensing often fails to distinguish a secure grasp from a light touch across varied objects and people, so the robot instead performs deliberate small movements to gather force data. A Bayesian model then treats the contact state as a distribution over piecewise-linear mappings from those motions to the forces a human applies. This lets the robot release the object only when multi-directional forces confirm a firm hold. A reader would care because reliable handovers are basic for any robot that assists in homes, hospitals, or factories. If the approach holds, robots can operate safely without constant human oversight or object-specific tuning.

Core claim

We propose an active sensing approach for robust handovers: the robot applies information-gathering motions and senses the resulting human-applied forces to infer the contact state. A firm grasp produces forces in multiple directions, while an accidental touch does not. To capture this distinction, we model the contact state with a Bayesian linear model: a distribution over piecewise-linear mappings from robot motions to human-applied forces. This model enables firm grasp detection and active information gathering. In experiments with 12 participants and 30 diverse rigid objects, our method achieved a 97.5% success rate -- over 30% higher than two common baselines.

What carries the argument

Bayesian linear model over piecewise-linear mappings from robot motions to human-applied forces, which separates firm multi-directional grasp responses from incidental single-direction touches and guides further motions for more data.

If this is right

  • The robot releases objects only after confirming multi-directional forces, avoiding drops from premature release or failed handovers from over-caution.
  • The same active-motion-plus-Bayesian-model pipeline works across 30 different rigid objects without per-object tuning.
  • Success reaches 97.5 percent, more than 30 percent above passive-sensing baselines, in tests with 12 people.
  • Information-gathering motions can be chosen on the fly to reduce uncertainty about the contact state.

Where Pith is reading between the lines

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

  • The method could be tested in settings where humans move while receiving the object, to check whether the piecewise-linear assumption still separates grasp from touch.
  • Similar active sensing might help other contact-rich tasks such as collaborative lifting or tool passing.
  • If the model is retrained online from new force data, it could adapt to soft or deformable objects not studied here.

Load-bearing premise

The distinction between firm grasp and incidental touch can be reliably captured by a Bayesian linear model over piecewise-linear mappings and will generalize across diverse rigid objects and human behaviors.

What would settle it

Running the system on a new collection of objects and participants and finding that incidental touches frequently produce multi-directional force patterns that the model labels as firm grasps.

Figures

Figures reproduced from arXiv: 2605.04610 by David Hsu, Linfeng Li, Lin Shao.

Figure 1
Figure 1. Figure 1: (a) We use a wrist-mounted force-torque (FT) sensor to sense view at source ↗
Figure 2
Figure 2. Figure 2: Our method centers on a human-object contact state represented view at source ↗
Figure 3
Figure 3. Figure 3: Contact states modeled as probabilistic piecewise-linear mapping view at source ↗
Figure 4
Figure 4. Figure 4: We selected 30 types of objects from the YCB dataset [7] and a human study [8]. These objects span a broad range of shapes and weights. view at source ↗
Figure 5
Figure 5. Figure 5: Hardware setup. The ATI Gamma sensor and Franka Hand are view at source ↗
Figure 6
Figure 6. Figure 6: Success rates for each method across 12 participants; error bars view at source ↗
read the original abstract

Robot-to-human object handover is an essential skill for robot assistants, from serving drinks at home to passing surgical tools in the operating room. We expect robots to perform handover robustly -- to release the object only after a firm human grasp while ignoring incidental touches. Existing passive-sensing methods struggle to generalize across diverse objects and human behaviors, as they lack informative perturbations to disambiguate different contact conditions, such as firm grasp versus incidental touch. We propose an active sensing approach for robust handovers: the robot applies information-gathering motions and senses the resulting human-applied forces to infer the contact state. A firm grasp produces forces in multiple directions, while an accidental touch does not. To capture this distinction, we model the contact state with a Bayesian linear model: a distribution over piecewise-linear mappings from robot motions to human-applied forces. This model enables firm grasp detection and active information gathering. In experiments with 12 participants and 30 diverse rigid objects, our method achieved a 97.5% success rate -- over 30% higher than two common baselines.

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 / 2 minor

Summary. The manuscript proposes an active contact sensing approach for robot-to-human object handovers. The robot performs information-gathering motions and senses resulting human forces, modeling contact state via a Bayesian linear model over piecewise-linear mappings from motions to forces. Firm grasps are distinguished by multi-directional force production (versus incidental touches), enabling detection and active sensing. Experiments with 12 participants and 30 diverse rigid objects report a 97.5% success rate, over 30% higher than two common baselines.

Significance. If the central results hold, the work offers a principled active-sensing strategy that could improve robustness and generalization in human-robot handovers compared to passive methods, with relevance to assistive and collaborative robotics. The Bayesian formulation for contact-state inference and information gathering is a constructive element that provides uncertainty-aware decision making.

major comments (2)
  1. [Abstract and §4] Abstract and §4 (Experiments): The 97.5% success rate and >30% improvement over baselines are reported without details on trial counts per condition, statistical tests, variance, baseline implementations, or failure-mode analysis, which are required to evaluate whether the quantitative claim is load-bearing and reproducible.
  2. [§3] §3 (Method): The Bayesian linear model over piecewise-linear mappings is defined to capture the multi-directional force signature of firm grasps versus incidental touch; however, no ablation or sensitivity analysis is provided to test whether this signature remains necessary and sufficient when humans employ stable uni-directional grips (due to object geometry, friction, or posture), which directly affects the posterior inference and active-sensing policy.
minor comments (2)
  1. [Abstract] The abstract refers to 'two common baselines' without naming them; the main text should explicitly identify and describe the baseline algorithms for clarity and reproducibility.
  2. [§3] Notation for the piecewise-linear mappings and the Bayesian update equations could be introduced with a small illustrative example or diagram to aid readers unfamiliar with the contact-state representation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback, which highlights important aspects for strengthening the clarity and robustness of our claims. We address each major comment below and will revise the manuscript to incorporate the suggested improvements where feasible.

read point-by-point responses

  1. Referee: [Abstract and §4] Abstract and §4 (Experiments): The 97.5% success rate and >30% improvement over baselines are reported without details on trial counts per condition, statistical tests, variance, baseline implementations, or failure-mode analysis, which are required to evaluate whether the quantitative claim is load-bearing and reproducible.

    Authors: We agree that the current presentation of results in the abstract and §4 lacks sufficient detail for full assessment of reproducibility and statistical validity. In the revised manuscript, we will expand §4 to explicitly report the total number of trials (12 participants × 30 objects × conditions), per-condition trial counts, standard deviations or confidence intervals, the precise implementations of the two baseline methods, appropriate statistical tests (e.g., paired t-tests with p-values), and a categorized failure-mode analysis of the unsuccessful cases. These additions will make the quantitative claims more transparent and load-bearing. revision: yes

  2. Referee: [§3] §3 (Method): The Bayesian linear model over piecewise-linear mappings is defined to capture the multi-directional force signature of firm grasps versus incidental touch; however, no ablation or sensitivity analysis is provided to test whether this signature remains necessary and sufficient when humans employ stable uni-directional grips (due to object geometry, friction, or posture), which directly affects the posterior inference and active-sensing policy.

    Authors: The concern is well-taken: the model's reliance on multi-directional force production could be sensitive to cases where humans apply stable uni-directional forces due to object shape or grip posture. Our experiments already span 30 objects with varied geometries and 12 participants with natural behaviors, and the active policy is designed to elicit informative multi-directional responses. However, we did not perform an explicit ablation isolating uni-directional grip subsets. In revision, we will add a sensitivity analysis (e.g., post-hoc examination of force directionality in successful vs. edge cases or a simulated uni-directional perturbation study) and discuss its implications for the posterior and policy. If new data collection is required, we will note this as a limitation. revision: partial

Circularity Check

0 steps flagged

No circularity: experimental results independent of any self-referential derivation

full rationale

The paper presents a proposed active sensing method that models contact states via a Bayesian linear model over piecewise-linear mappings, then evaluates it through physical experiments on 30 objects with 12 participants. The 97.5% success rate is reported as an empirical outcome compared against baselines, with no equations, predictions, or first-principles derivations shown that reduce the performance metric or model parameters to fitted inputs or self-citations by construction. The distinction between firm grasp and incidental touch is explicitly posited as a modeling choice rather than derived from prior results, and no load-bearing uniqueness theorems or ansatzes from the authors' own prior work are invoked to force the approach.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that force patterns from robot motions reliably separate firm grasp from incidental touch, plus the modeling choice of Bayesian linear regression over piecewise-linear mappings.

axioms (1)
  • domain assumption A firm grasp produces forces in multiple directions while an accidental touch does not.
    Explicitly stated in the abstract as the physical basis for the sensing strategy.

pith-pipeline@v0.9.0 · 5481 in / 1116 out tokens · 60687 ms · 2026-05-08T16:07:55.532142+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

35 extracted references · 35 canonical work pages

  1. [1]

    Experimental testing of the CogLaboration prototype system for fluent human-robot object handover interactions,

    A. Koene, S. Endo, A. Remazeilles, M. Prada, and A. M. Wing, “Experimental testing of the CogLaboration prototype system for fluent human-robot object handover interactions,” inIEEE Int. Symp. on Robot and Human Interactive Communication, 2014

    work page 2014

  2. [2]

    Robotic scrub nurse to anticipate surgical instruments based on real-time laparoscopic video analysis,

    L. Wagner, S. Jourdan, L. Mayer, C. M ¨uller, L. Bernhard, S. Kolb, F. Harb, A. Jell, M. Berlet, H. Feussner,et al., “Robotic scrub nurse to anticipate surgical instruments based on real-time laparoscopic video analysis,”Communications Medicine, vol. 4, no. 1, p. 156, 2024

    work page 2024

  3. [3]

    Towards autonomous robotic butlers: Lessons learned with the PR2,

    J. Bohren, R. B. Rusu, E. Gil Jones, E. Marder-Eppstein, C. Pantofaru, M. Wise, L. M ¨osenlechner, W. Meeussen, and S. Holzer, “Towards autonomous robotic butlers: Lessons learned with the PR2,” inProc. IEEE Int. Conf. on Robotics & Automation, 2011

    work page 2011

  4. [4]

    A human-inspired object handover controller,

    W. P. Chan, C. A. Parker, H. M. V . der Loos, and E. A. Croft, “A human-inspired object handover controller,”Int. J. Robotics Research, vol. 32, no. 8, pp. 971–983, 2013

    work page 2013

  5. [5]

    A human- inspired controller for fluid human-robot handovers,

    J. R. Medina, F. Duvallet, M. Karnam, and A. Billard, “A human- inspired controller for fluid human-robot handovers,” inProc. IEEE- RAS Int. Conf. on Humanoid Robots, 2016, pp. 324–331

    work page 2016

  6. [6]

    Learning dynamic robot-to- human object handover from human feedback,

    A. Kupcsik, D. Hsu, and W. S. Lee, “Learning dynamic robot-to- human object handover from human feedback,” inProc. Int. Symp. on Robotics Research, 2018

    work page 2018

  7. [7]

    The YCB object and model set: Towards common benchmarks for manipulation research,

    B. Calli, A. Singh, A. Walsman, S. Srinivasa, P. Abbeel, and A. M. Dollar, “The YCB object and model set: Towards common benchmarks for manipulation research,” inProc. IEEE Int. Conf. on Robotics & Automation, 2015

    work page 2015

  8. [8]

    A list of household objects for robotic retrieval prioritized by people with als,

    Y . S. Choi, T. Deyle, T. Chen, J. D. Glass, and C. C. Kemp, “A list of household objects for robotic retrieval prioritized by people with als,” inProc. IEEE Int. Conf. on Rehabilitation Robotics, 2009

    work page 2009

  9. [9]

    Object handovers: A review for robotics,

    V . Ortenzi, A. Cosgun, T. Pardi, W. P. Chan, E. Croft, and D. Kuli ´c, “Object handovers: A review for robotics,”IEEE Trans. on Robotics, pp. 1–19, 2021

    work page 2021

  10. [10]

    Toward seamless human-robot handovers,

    K. Strabala, M. K. Lee, A. Dragan, J. Forlizzi, S. S. Srinivasa, M. Cak- mak, and V . Micelli, “Toward seamless human-robot handovers,” Journal of Human-Robot Interaction, vol. 2, no. 1, p. 112–132, feb 2013

    work page 2013

  11. [11]

    RFID-guided robots for pervasive automation,

    T. Deyle, H. Nguyen, M. Reynolds, and C. Kemp, “RFID-guided robots for pervasive automation,”IEEE Pervasive Computing, vol. 9, no. 2, pp. 37–45, 2010

    work page 2010

  12. [12]

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

    A. G. Egu ´ıluz, I. Ra˜n´o, S. A. Coleman, and T. M. McGinnity, “Reliable object handover through tactile force sensing and effort control in the shadow robot hand,” inProc. IEEE Int. Conf. on Robotics & Automation, 2017

    work page 2017

  13. [13]

    The effects of proactive release behaviors during human-robot handovers,

    Z. Han and H. Yanco, “The effects of proactive release behaviors during human-robot handovers,” inProc. ACM/IEEE Int. Conf. on Human-Robot Interaction, 2019

    work page 2019

  14. [14]

    Delivery by hand between human and robot based on fingertip force-torque information,

    K. Nagata, Y . Oosaki, M. Kakikura, and H. Tsukune, “Delivery by hand between human and robot based on fingertip force-torque information,” inProc. IEEE/RSJ Int. Conf. on Intelligent Robots & Systems, 1998

    work page 1998

  15. [15]

    Bidirectional human-robot bimanual handover of big planar object with vertical posture,

    W. He, J. Li, Z. Yan, and F. Chen, “Bidirectional human-robot bimanual handover of big planar object with vertical posture,”IEEE Trans. on Automatic Science and Engineering, vol. 19, no. 2, pp. 1180– 1191, 2022

    work page 2022

  16. [16]

    Compliant blind handover control for human-robot collaboration,

    D. Ferrari, A. Pupa, and C. Secchi, “Compliant blind handover control for human-robot collaboration,” inProc. IEEE/RSJ Int. Conf. on Intelligent Robots & Systems, 2024

    work page 2024

  17. [17]

    A fail-safe object handover controller,

    S. Parastegari, E. Noohi, B. Abbasi, and M. ˇZefran, “A fail-safe object handover controller,” inProc. IEEE Int. Conf. on Robotics & Automation, 2016

    work page 2016

  18. [18]

    Joint action understanding improves robot-to-human object han- dover,

    E. C. Grigore, K. Eder, A. G. Pipe, C. Melhuish, and U. Leonards, “Joint action understanding improves robot-to-human object han- dover,” inProc. IEEE/RSJ Int. Conf. on Intelligent Robots & Systems, 2013

    work page 2013

  19. [19]

    Human inspired grip-release technique for robot-human handovers,

    P. Khanna, M. Bj ¨orkman, and C. Smith, “Human inspired grip-release technique for robot-human handovers,” inProc. IEEE-RAS Int. Conf. on Humanoid Robots, 2022

    work page 2022

  20. [20]

    Environment estimation for enhanced impedance control,

    L. Love and W. Book, “Environment estimation for enhanced impedance control,” inProc. IEEE Int. Conf. on Robotics & Automa- tion, 1995

    work page 1995

  21. [21]

    Contact stiffness and damping estimation for robotic systems,

    D. Erickson, M. Weber, and I. Sharf, “Contact stiffness and damping estimation for robotic systems,”Int. J. Robotics Research, vol. 22, no. 1, pp. 41–57, 2003

    work page 2003

  22. [22]

    Contact impedance estimation for robotic systems,

    N. Diolaiti, C. Melchiorri, and S. Stramigioli, “Contact impedance estimation for robotic systems,”IEEE Trans. on Robotics, vol. 21, no. 5, pp. 925–935, 2005

    work page 2005

  23. [23]

    A new method for online parameter estimation of hunt-crossley environment dynamic models,

    A. Haddadi and K. Hashtrudi-Zaad, “A new method for online parameter estimation of hunt-crossley environment dynamic models,” inProc. IEEE/RSJ Int. Conf. on Intelligent Robots & Systems, 2008, pp. 981–986

    work page 2008

  24. [24]

    Learning physical collaborative robot behaviors from human demon- strations,

    L. Rozo, S. Calinon, D. G. Caldwell, P. Jim ´enez, and C. Torras, “Learning physical collaborative robot behaviors from human demon- strations,”IEEE Trans. on Robotics, vol. 32, no. 3, pp. 513–527, 2016

    work page 2016

  25. [25]

    Sensorless optimal interaction control exploiting environment stiffness estimation,

    L. Roveda, A. A. Shahid, N. Iannacci, and D. Piga, “Sensorless optimal interaction control exploiting environment stiffness estimation,”IEEE Trans. on Control System Technology, vol. 30, no. 1, pp. 218–233, 2022

    work page 2022

  26. [26]

    Passive interaction control with dynam- ical systems,

    K. Kronander and A. Billard, “Passive interaction control with dynam- ical systems,”IEEE Robotics & Automation Letters, vol. 1, no. 1, pp. 106–113, 2016

    work page 2016

  27. [27]

    C. M. Bishop and N. M. Nasrabadi,Pattern Recognition and Machine Learning, 2006

    work page 2006

  28. [28]

    K. M. Lynch and F. C. Park,Modern Robotics, 2017

    work page 2017

  29. [29]

    Ben-Tal, L

    A. Ben-Tal, L. El Ghaoui, and A. Nemirovski,Robust Optimization, 2009

    work page 2009

  30. [30]

    S. Boyd, S. P. Boyd, and L. Vandenberghe,Convex Optimization, 2004

    work page 2004

  31. [31]

    Thrun, W

    S. Thrun, W. Burgard, and D. Fox,Probabilistic Robotics, 2005

    work page 2005

  32. [32]

    NASA-task load index (NASA-TLX); 20 years later,

    S. G. Hart, “NASA-task load index (NASA-TLX); 20 years later,” inProceedings of the human factors and ergonomics society annual meeting, 2006

    work page 2006

  33. [33]

    SUS-a quick and dirty usability scale,

    J. Brookeet al., “SUS-a quick and dirty usability scale,”Usability evaluation in industry, vol. 189, no. 194, pp. 4–7, 1996

    work page 1996

  34. [34]

    The measurement of observer agreement for categorical data,

    J. R. Landis and G. G. Koch, “The measurement of observer agreement for categorical data,”Biometrics, pp. 159–174, 1977

    work page 1977

  35. [35]

    Sensorless robot collision detection and hybrid force/motion control,

    A. de Luca and R. Mattone, “Sensorless robot collision detection and hybrid force/motion control,” inProc. IEEE Int. Conf. on Robotics & Automation, 2005

    work page 2005