PaperlayerView exact revision on arXiv

Source-linked AI summary

TacClip: a clip-on sensor measures dynamic contact forces without covering the fingerpads

Yuqian Ye, Hao Li, Jingxi Xu, Haojun Feng, Seongheon Hong, Mark R. Cutkosky

arXiv:2609.08214v1cs.RO

TL;DR

Contact-rich manipulation depends on force and tactile events, but many sensing approaches encumber the fingerpad and can interfere with bare-finger cues. TacClip uses a clip-on FBG to measure fingertip deformation while leaving the pad exposed; tests report force-estimation errors below 0.5 N and capture dynamic tactile signals during exploration.

  • Problem

    Contact-rich manipulation depends on contact timing, force modulation, and tactile events, while many sensing approaches encumber the fingerpad or interfere with bare-finger cues.

    Source: §I · p. 1

  • Method

    TacClip clips around the fingertip and uses an embedded FBG to convert contact-induced deformation into wavelength shifts while leaving the fingerpad exposed.

    Source: §I · p. 1

  • Results

    Typical force-estimation error is below 0.5 N over the tested range, and TacClip captures vibrations and dynamic events during exploratory sliding.

    Source: A. Quasi-static Force Response · p. 3, §VI · p. 7

  • Takeaways & Limitations

    TacClip adds a dynamic contact channel while preserving exposed fingerpads, and its photonic sensing is unaffected by wet conditions and electromagnetic interference.

    Source: §VI · p. 7

    TacClip does not decouple fx, fy, and fz; its overall-load estimates depend on finger geometry, tissue stiffness, fit, preload, and temperature.

    Source: §VI · p. 7

Abstract

from arXiv · show

TacClip is a minimally encumbering wearable device for recording fingertip deformation caused by contact forces and vibrations. It can be combined with vision- or glove-based hand tracking systems that leave the fingertips uncovered and provides a measure of dynamic contact interactions, while leaving the finger pads exposed so that the user retains natural sensitivity to texture, friction, temperature, and fine surface features. The signal is produced by a Fiber Bragg Grating (FBG) embedded on a small plastic clip mounted over the fingernail. Optionally, for use with vision-based tracking, additional FBGs on polyimide strips can complement camera-based pose estimation. In finger pressing tests, TacClip estimates the force magnitude with typical errors below $0.5~\mathrm{N}$ over a $0$--$8~\mathrm{N}$ range. In tests of cloth handling and tape edge finding, we show that it captures the vibrations and dynamic events generated during exploratory sliding. With no electronics, TacClip can also be used submerged in water, while preserving bare finger contact.

I. INTRODUCTION

Contact-rich demonstrations need force and tactile information, but many sensing approaches encumber the fingerpad and can interfere with the bare-finger cues used during exploration. TacClip clips around the fingertip to measure contact-induced deformation while leaving the pad exposed.

I. INTRODUCTION

  • Contact-rich manipulation depends on when contact occurs, how force is modulated, and what tactile events arise during sliding, pressing, or grasping.Robot-learning demonstrations often emphasize visual observations and hand or tool trajectories.Source: §I · p. 1
  • Many fingertip sensors place material between the fingerpad and environment, restrict dexterity, or require bulky fixtures.These constraints matter for sensing bare-finger cues such as fabric ridges, transparent edges, and compliant surfaces.Source: §I · p. 1
  • TacClip clips around the fingertip; contact deforms the clip, and an embedded FBG converts that deformation into wavelength shifts.The fingerpad remains exposed to the environment.Source: §I · p. 1
  • Optional FBGs on a dorsal polyimide strip measure curvature to augment visual hand-pose tracking without a full glove.The strip can carry additional FBGs along the same optical fiber.Source: §I · p. 1, Figure 1 · p. 1
  • The paper characterizes quasi-static force estimation and dynamic vibration sensing, then demonstrates cloth-texture discrimination, tape-edge detection, and underwater force-response measurement.It also discusses integrating the optional curvature strip with visual hand tracking.Source: §I · p. 1

A. Tactile and Force Information for Robot Learning

Robot-learning systems increasingly collect human demonstrations, but visual and motion-capture pipelines can omit force and touch information. TacClip is positioned among approaches that seek richer contact sensing while preserving bare-finger perception.

A. Tactile and Force Information for Robot Learning

  • Visual, pose, and motion-capture pipelines support scalable demonstration capture, while force- and touch-centered systems show the value of contact information for transferring human strategies to robot policies.TacClip targets bare-finger demonstrations without a handheld gripper, thick glove, or sensor layer under the pad.Source: A. Tactile and Force Information for… · p. 2

B. Wearable Fingertip and Glove Sensors

  • Gloves can record large-area tactile patterns, but glove layers may attenuate bare-skin perception or alter contact friction; ThimbleSense instruments the contact surface.TacClip instead measures deformation around the fingertip while leaving the fingerpad uncovered.Source: B. Wearable Fingertip and Glove Sensors · p. 2

C. FBG-Based Tactile and Hand-Posture Sensing

D. Dynamic Touch Sensing

Dynamic touch sensing uses vibration and skin acceleration to study texture and slip, while TacClip records contact-induced vibration during natural exploratory motion. Its clip bends with fingertip expansion, shifting the FBG wavelength; optional dorsal FBGs provide curvature signals to complement visual pose tracking.

D. Dynamic Touch Sensing

  • Skin acceleration and vibration contain information about slip and surface texture, and fabric exploration can use tactile feedback to distinguish fine structures difficult to infer from vision.Lateral motion is a characteristic human exploratory procedure for texture assessment.Source: D. Dynamic Touch Sensing · p. 2

III. MECHANICAL DESIGN OF TACCLIP

  • The 3D-printed clip slides over the fingertip without adhesive and leaves the pad uncovered; fit and preload affect the wavelength baseline and force calibration.Three nominal inner diameters cover the tested fingertip-width range, and visible clip motion requires baseline reset or recalibration.Source: §III · p. 2
  • Fingertip tissue bulges laterally under load and bends the clip, straining the fiber and shifting its reflected Bragg wavelength.The geometry is sensitive to loads that produce lateral fingertip expansion; Fig. 2 depicts the clip and simulated strut strain.Source: §III · pp. 2–3, Figure 2 · p. 3
  • Optional FBGs on a dorsal polyimide strip measure local curvature and can provide flexion information during occlusions when fused with visual skeleton estimates.These signals do not replace full hand-pose tracking.Source: §III · p. 3

IV. CHARACTERIZATION OF TACCLIP

TacClip estimates contact-force magnitude from fingertip deformation, which is nonlinear and affected by viscoelastic history. Its characterization covers force estimation, cross-user calibration, and dynamic response.

A. Quasi-static Force Response

  • Viscoelastic fingertip tissue produces hysteresis and relaxation, so transient deformation can persist after a rapid load-and-unload event.The main static response is monotonic but nonlinear, motivating data-driven calibration.Source: A. Quasi-static Force Response · p. 3
  • Typical force-estimation error is below 0.5 N in the tested range, using an MLP that maps the TacClip signal and recent history to force estimates.The force/torque sensor provides ground-truth components; Fig. 3 compares learned estimates with those measurements.Source: A. Quasi-static Force Response · p. 3
  • The single-FBG prototype is best interpreted as an estimator of contact-force magnitude, not normal force, because wavelength shift follows overall fingertip deformation more closely.With shear added while normal force stays nearly constant, total force magnitude increases.Source: A. Quasi-static Force Response · p. 3
  • The dynamic-response characterization uses a swept-frequency stimulus and examines the wavelength-shift response relative to applied force.Fig. 4 depicts a muscle-lever chirp setup and a Bode plot with coherence.Source: Figure 4 · p. 4
  • Across five users, normalized quasi-static wavelength-force curves remain highly consistent despite differences in finger size and initial preloading.The transferred baseline MLP uses a brief calibration step to account for individual tissue hysteresis.Source: A. Quasi-static Force Response · p. 4

B. Dynamic Tactile Sensing

TacClip’s dynamic response is characterized with swept-frequency loading and wavelength-shift measurements. Its response is relatively stable at lower frequencies, while fingertip mechanics attenuate rapid vibrations.

B. Dynamic Tactile Sensing

  • The interrogator samples at 2 kHz, but soft fingertip tissue attenuates high-frequency vibrations before they produce measurable clip deformation.The stated sampling rate is sufficient to capture vibration-rich events during sliding, tapping, and edge crossing.Source: B. Dynamic Tactile Sensing · p. 4
  • A muscle-lever system applied an approximately constant-amplitude chirp to the fingerpad, with frequency increasing from 0 to 100 Hz.TacClip was worn during the test to characterize the response under realistic wearing conditions.Source: B. Dynamic Tactile Sensing · p. 4
  • Perpendicular sliding on corduroy produces more regular and intense periodic vibrations than sliding parallel to the ridges.Fig. 5 compares raw and high-pass filtered signals and their power spectral densities for the two sliding directions.Source: Figure 5 · p. 4
  • FFT analysis quantified the transfer function from applied force to resulting FBG wavelength shift.The frequency-response figure presents the wavelength-shift response divided by applied force.Source: B. Dynamic Tactile Sensing · p. 4
  • The magnitude and phase response remain relatively stable from near-DC to roughly 20–30 Hz, with low phase lag and high coherence; above this range, coherence decreases and response becomes less consistent.A measurable response persists above that range, supporting capture of pressing and holding force variations and a portion of sliding vibration.Source: B. Dynamic Tactile Sensing · p. 4

V. APPLICATION EXPERIMENTS

The application experiments evaluate tactile tasks where bare-finger cues are useful and vision alone can be ambiguous. They include fabric-direction sensing, transparent tape-edge detection, and underwater force measurement.

V. APPLICATION EXPERIMENTS

  • The application experiments evaluate fabric-direction sensing, transparent tape-edge detection, and contact-force measurement underwater.Source: §V · p. 4
  • The selected tasks are ones in which bare-finger tactile cues are useful and vision alone can be ambiguous.Source: §V · p. 4
  • The experiments follow the characterization of TacClip’s contact-force magnitude trends and dynamic tactile-event measurements.Source: §V · p. 4

A. Cloth Alignment

TacClip detects cloth-ridge direction during fingertip sliding by capturing direction-dependent vibrations. Perpendicular motion produces larger wavelength variations than parallel motion.

A. Cloth Alignment

  • The test compares fingertip sliding perpendicular and parallel to the corduroy ridges.The direction-dependent vibration cues could help distinguish cloth orientation during exploratory motion.Source: A. Cloth Alignment · pp. 4–5
  • Perpendicular sliding over corduroy ridges produces approximately ±0.004 nm wavelength variations, versus approximately ±0.001 nm for parallel sliding.The signals were recorded under comparable applied forces.Source: A. Cloth Alignment · p. 5
  • Power spectral density analysis further distinguishes the two sliding directions in the frequency domain.The signals were also high-pass filtered at 30 Hz to remove baseline contact-force effects.Source: A. Cloth Alignment · p. 5

B. Transparent Tape Edge Detection

TacClip detects transparent tape boundaries from repeatable signal peaks, with detection improving at higher pressing forces. After temperature calibration, its spring-compression response in air and water is not significantly different, supporting wet use when temperature is compensated.

B. Transparent Tape Edge Detection

  • As pressing force increases, tape-boundary detection accuracy and precision improve, and the interval range narrows.The authors attribute this to higher signal-to-noise ratio at greater contact forces; Table I summarizes detection results under different pressing forces.Source: B. Transparent Tape Edge Detection · p. 5
  • Repeatable signal peaks occur as the fingertip crosses the tape boundary, and increasing pressing force raises peak amplitude and signal-to-noise ratio.The evaluation used 78 µm packing tape on a turntable rotating at 45 RPM, corresponding to 21.25 cm/s.Source: B. Transparent Tape Edge Detection · p. 5

C. Underwater Force Response

  • Spring length measured from side-view video served as a proxy for applied force in the compression test.The estimated spring-length measurement accuracy was approximately ±1.0 mm per reading.Source: C. Underwater Force Response · p. 5
  • After temperature calibration, the air and water wavelength-shift curves are not significantly different.The authors suggest wet or underwater tactile data collection when temperature is compensated.Source: Figure 7 · p. 6, C. Underwater Force Response · p. 6

D. Surgical Palpation

In tissue-phantom palpation, TacClip records high-frequency signal variation over a vascular region, unlike the smooth low-frequency variation over a non-vascular region. The dynamic signals are associated with subcutaneous features while leaving tactile sensitivity unobstructed.

D. Surgical Palpation

  • TacClip leaves tactile sensitivity unobstructed for detecting changes in compliance and texture relevant to palpation.The paper frames these cues as relevant to tasks such as locating blood vessels or subcutaneous masses.Source: D. Surgical Palpation · p. 6
  • Sliding over the vascular region produces high-frequency variations superimposed on the baseline low-frequency motion profile.The non-vascular region produces only smooth, low-frequency variation from sliding.Source: D. Surgical Palpation · p. 6
  • The spectral difference between vascular and non-vascular regions indicates dynamic signals associated with subcutaneous features during palpation.The demonstration used a tissue phantom.Source: D. Surgical Palpation · p. 6, Figure 8 · p. 6

E. Augmenting Hand Pose Information

TacClip’s polyimide-strip FBGs measure finger flexion to augment camera-based hand-pose tracking. In grasping tests, vision-FBG fusion improves dimensional accuracy and tracking stability over vision-only tracking.

E. Augmenting Hand Pose Information

  • FBGs along polyimide strips measure local curvature, providing finger-flexion information that can augment vision-based pose estimates, especially under partial occlusion.The measurement does not fully capture finger pose.Source: E. Augmenting Hand Pose Information · p. 6
  • The fusion framework trains an MLP from FBG strain profiles and motion-capture ground-truth joint angles, then combines FBG flexion with camera abduction estimates when visual confidence falls.The framework was evaluated for grasping with the thumb, index, and middle fingers.Source: E. Augmenting Hand Pose Information · pp. 6–7
  • The vision-FBG fusion approach improves dimensional accuracy and tracking stability over vision-only tracking in grasping examples.The comparison covers a cylinder, small box, and thin card of known dimensions; the table reports hand-pose estimation accuracy during object grasping.Source: Table II · p. 7, E. Augmenting Hand Pose Information · p. 7

VI. DISCUSSION AND CONCLUSION

TacClip adds a bare-finger contact channel that estimates overall contact-load magnitude and captures dynamic tactile events. It complements rather than replaces other hand-sensing modalities, and its estimates remain bounded by the tested conditions.

VI. DISCUSSION AND CONCLUSION

  • With supervised calibration, TacClip estimates contact-force magnitude with typical errors below 0.5 N over the tested 0–8 N range.The prototype’s 2 kHz signal also captures dynamic events and vibrations during texture exploration.Source: §VI · p. 7
  • Its 2 kHz signal captures vibrations and dynamic tactile phenomena that other modalities may miss.The sensor is intended to add a contact channel, not replace vision, pose tracking, force/torque sensing, or distributed tactile gloves.Source: §VI · p. 7
  • Photonic sensing leaves TacClip unaffected by wet conditions and electromagnetic interference.Source: §VI · p. 7
  • TacClip estimates overall contact-load magnitude under the tested conditions; it does not decouple fx, fy, and fz.Response depends on finger geometry, tissue stiffness, fit, preload, and temperature, motivating rapid per-user calibration.Source: §VI · p. 7
  • A snug, repeatable fit is required because clip position and preload affect calibration.The clip head slides over the fingertip, needs no adhesive, and can be fabricated in different sizes.Source: §VI · p. 7
  • At prototype scale, each sensor head costs approximately $20 per finger, plus $80 for the dorsal wristband, excluding the optical interrogator.The interrogator is currently costly compared with strain-gauge amplifiers and analog-to-digital conversion, though its cost is dropping as photonic sensing grows more popular.Source: §VI · pp. 7–8

Related papers

Loading 2609.08214v1…