Source-linked AI summary
Tactile Sensing Array for Multi-Phalanx Sensing in Humanoid Hands
Neel Adwani, Muhaiminul Islam Akash, Rituja Bhattacharya, Cong Wang
TL;DR
Humanoid hands need tactile feedback across the finger because vision and proprioception alone cannot reliably localize contact or regulate a grasp. The paper presents a low-cost Velostat array with distributed phalanx sensing and tests its hysteresis, contact patterns, and recovery. The gap-and-bump structure reduces recovery time relative to a flush-contact baseline.
Problem
Humanoid hands need tactile feedback for grasping and manipulation because vision can be occluded by the hand and joint angles alone do not reveal contact.
Source: §I · p. 1Method
Results
Recovery time fell 74% with the gap-and-bump design versus the flush-contact baseline, from 4.6 ± 1.4 s to 1.2 ± 0.7 s.
Source: B. Post-Release Recovery Time · p. 3Takeaways & Limitations
The sensor produces distinct fingertip activation patterns for different contact geometries and registers contact across all three phalanges during a grip.
Source: C. Contact Discrimination · p. 4, §V · p. 4Evaluation used cube and cylinder objects; blind tactile-only manipulation and generalization to a wider range of object shapes and sizes remain future tests.
Source: §V · p. 4
Abstract
from arXiv · showhide
Humanoid hands require tactile feedback across the whole finger, not just the fingertip, to grasp and manipulate objects properly. Vision and proprioception alone cannot reliably provide this information, particularly when the hand's own fingers occlude the camera's view of the grasp. We present a low-cost tactile array for a humanoid finger, made from Velostat and conductive tape. The fingertip carries seven contact points including a 2x3 matrix wrapped across its front, left, and right faces, and a separate contact point at the tip. The proximal and middle phalanges each carry a single front-facing contact line. We measure the sensor's hysteresis and recovery time after release, through repeated loading and press-release tests. We also test a compliant, 3D-printed contact structure with a gap and a bump, inspired by similar designs in prior work, and show it cuts recovery time by 74% compared to a flush-contact baseline. We then show the sensor can produce distinct activation patterns for different contact geometries (flat, edge, corner) at the fingertip, and that it registers contact across all three phalanges during a grip. Finally, we discuss the limits of our fabrication changes and point to software-based compensation as a promising way to more directly fix the remaining hysteresis in the future.
I. INTRODUCTION
Reliable manipulation in unstructured environments calls for tactile feedback because vision and proprioception have important limits. The design distributes sensing according to the different needs of the fingertip and other phalanges.
I. INTRODUCTION
- Closed-loop manipulation uses tactile feedback to regulate grip force, localize contact, detect slip, and respond to unexpected contact.Vision can be occluded by the hand or affected by lighting, while joint angles alone do not reveal whether or how hard the hand touches an object.Source: §I · p. 1
- The fingertip needs richer tri-directional sensing, whereas front-facing sensing is adequate for proximal finger regions.Uniformly high-resolution sensing across every phalanx would spend wiring and effort where it is not needed; insufficient fingertip sensing risks losing contact information for precise manipulation.Source: §I · p. 1
- Most existing tactile sensors cover only the fingertip front, omitting lateral contact during in-hand manipulation and power grasps.The passage identifies this missing lateral coverage as a design gap.Source: §I · p. 1
- The paper presents a low-cost Velostat matrix array with richer fingertip sensing and front-only sensing on the proximal and middle phalanges.It characterizes hysteresis and post-release relaxation, tests a 3D-printed gap-and-bump structure against a flush baseline, and identifies software compensation as a future direction for residual hysteresis.Source: §I · p. 1
II. RELATED WORK
Existing tactile approaches trade coverage against cost, integration complexity, or fabrication demands, and Velostat-based designs have left multi-phalanx coverage and hysteresis behavior under-characterized. This work tests an accessible contact structure for recovery time in a multi-contact fingertip design.
II. RELATED WORK
- High-resolution vision-, biomimetic-, and magnetic-skin sensors often trade coverage against cost and integration complexity and are commonly limited to the fingertip.DexSkin offers high-coverage sensing across finger surfaces and the palm, but uses more complex, application-specific fabrication than a simple resistive matrix.Source: §II · pp. 1–2
- Prior Velostat designs cover fin-shaped gripper fingers or a fingertip front, but do not characterize hysteresis and recovery time together with curved multi-phalanx sensing.3D-ViTac does not characterize hysteresis or relaxation; LeapTac reports spatial resolution and minimum force threshold without hysteresis or recovery time.Source: §II · p. 2
- Prior hardware interventions reduce hysteresis-related error or achieve sub-100 ms response and recovery, but accessible multi-phalanx fabrication interventions remain unevaluated.The cited interventions include magnet prestressing,, stacked Velostat layers, and a precision-molded gap-and-bump mechanism; the paper tests a compliant 3D-printed spacer and bump for post-release recovery time.Source: §II · p. 2
A. Design Rationale
Sensing needs vary across the finger: the fingertip benefits from front-and-side coverage and a dedicated tip point, while other phalanges provide complementary grasp information. The design also adapts a gap-and-bump mechanism to accessible fabrication.
A. Design Rationale
- Middle and proximal phalanges provide complementary information during grasping and manipulation, and some power grasps require sensing on those regions.The cited simulation work maps phalange contributions across object shapes and sizes; the passage states that without middle and proximal sensing, a hand cannot distinguish a precision grasp from a prehensile one.Source: A. Design Rationale · p. 2
- Fingertip contact can occur on palmar and lateral surfaces, while precision grasps often concentrate contact at the very tip.The design therefore separates a dedicated distal-tip contact point from front-and-lateral sensing.Source: A. Design Rationale · p. 2
- The contact mechanism adapts the gap-and-bump approach in using standard 3D printing and off-the-shelf materials instead of precision molding and custom microfabrication.The cited prior design reports sub-100 ms response and recovery; the paper's stated aim is fabrication without specialized equipment.Source: A. Design Rationale · p. 2
B. Electrode Layout and Signal Acquisition
The electrode layout places a wrapped matrix and separate tip point on the distal phalanx, with simpler front-facing lines on the middle and proximal phalanges. The matrix is multiplexed for signal acquisition.
B. Electrode Layout and Signal Acquisition
- The distal phalanx has seven contact points: a 2×3 matrix across front, left, and right faces, plus an independently wired point at the tip.The six matrix points share two row electrodes and five electrical wires; each proximal and middle phalanx has one front-facing contact line.Source: B. Electrode Layout and Signal… · p. 2
- The matrix is scanned column by column at 20 Hz, driving one row HIGH at a time and reading each column through a voltage divider.A pull-down resistor is sized empirically from Velostat's measured resistance range.Source: B. Electrode Layout and Signal… · p. 2
- The independent tip point and proximal and middle contact lines use the same voltage-divider principle without multiplexing.The passage states this is consistent with their role.Source: B. Electrode Layout and Signal… · p. 2
C. Fabrication Process
The distal phalanx stack combines layered electrodes, Velostat, a spacer, and a bump plate; proximal and middle phalanx sensors use simpler constructions.
C. Fabrication Process
- The distal stack places a Velostat layer between row and column electrodes, with a spacer and bump plate above it.The backing, spacer, and bump plates are 3D-printed in TPU; the spacer creates an air gap and the raised bumps concentrate contact force.Source: C. Fabrication Process · p. 3
- Assembly bonds two electrode rows and the tip electrode to the backing, covers non-electrode areas with Velostat, then adds the spacer, columns, and bump plate.The layers are bolted together.Source: C. Fabrication Process · p. 3
- The proximal and middle phalanx sensors have simpler constructions, and spacer and bump dimensions were selected for best recovery time.The selected dimensions are detailed in Section IV.Source: C. Fabrication Process · p. 3
IV. TESTING AND RESULTS
The sensor was tested by gradually loading and unloading a representative contact point while recording force and raw readings. The resulting curves differ between loading and unloading, confirming hysteresis.
A. Hysteresis Characterization
- Force was increased from 0 g to 50 g in 5 g increments, then decreased back to 0 g.Raw sensor values were recorded at multiple points along both paths while applied force was measured.Source: A. Hysteresis Characterization · p. 3
- The test recorded raw readings along both the loading and unloading paths.A single representative contact point was loaded using increments measured on a scale.Source: A. Hysteresis Characterization · p. 3
- The loading and unloading curves do not coincide, confirming hysteresis in the proposed sensor.At a given force, the reading differs depending on whether the sensor is being loaded or unloaded, consistent with Velostat’s reported visco-elastic behavior,.Source: A. Hysteresis Characterization · p. 3, Figure 4 · p. 3
B. Post-Release Recovery Time
A compliant TPU gap-and-bump structure shortened the sensor’s post-release recovery time compared with a flush-contact baseline.
B. Post-Release Recovery Time
- 74%: the gap-and-bump design reduced mean recovery time from 4.6 ± 1.4 s to 1.2 ± 0.7 s versus the flush-contact baseline.The comparison used five repeated trials.Source: B. Post-Release Recovery Time · p. 3
- The best-performing dimensions were a 0.4 mm spacer and a 0.6 mm bump height.Both the spacer and bump layers were fabricated from TPU, unlike the rigid-spacer/compliant-bump design in.Source: B. Post-Release Recovery Time · p. 3
- Recovery time measured how long the raw signal took to return within 20% of its pre-contact baseline after release.A known weight was applied to one contact point, held briefly, and then released.Source: B. Post-Release Recovery Time · p. 3
C. Contact Discrimination
The fingertip array produced different activation patterns across contact geometries, while a cylinder grip registered contact across the distal, middle, and proximal phalanges.
C. Contact Discrimination
- Tip, frontal, edge, corner, and lateral contacts produced distinct activation patterns across the seven distal contact points.Tip contact activated only the tip point; frontal contact spread across the matrix front, edge contact formed a line, corner contact activated one point, and lateral flat contact activated the right side.Source: C. Contact Discrimination · p. 4, Figure 5 · p. 4
- A cylinder grip activated the distal matrix and the middle and proximal contact lines simultaneously.This grip exercised the full sensing array.Source: C. Contact Discrimination · pp. 3–4
- The fingertip test used a cube and varied its contact orientation across five ways.The resulting Fig. 5 panels show activation across the seven distal-phalanx contact points.Source: C. Contact Discrimination · pp. 3–4
V. CONCLUSIONS AND FUTURE WORK
The low-cost Velostat array produces distinct fingertip patterns and registers contact across all three phalanges during a grip; its compliant gap-and-bump structure reduces recovery time relative to flush contact. Future work targets whole-hand deployment, broader testing, and kinematic-state-informed hysteresis correction.
V. CONCLUSIONS AND FUTURE WORK
- The sensor produces distinct patterns for different fingertip contact geometries and registers contact across all three phalanges during a grip.The compliant, 3D-printed gap-and-bump structure reduces recovery time compared with a flush-contact baseline.Source: §V · p. 4
- Extending the design beyond a single finger to the palm and all fingers is future work, with the aim of sharing tactile feedback during a grasp.Source: §V · p. 4
- Future evaluation will test blind grasping, more object shapes and sizes beyond the cube and cylinder, and kinematic-state inputs for hysteresis correction.The planned correction would use joint angle, joint velocity, and contact duration rather than relying on tactile signals alone.Source: §V · p. 4