Vision-based tactile sensors (VBTS) provide rich contact information for robotic manipulation, but existing designs can be hard to simplify and adapt to the size and constraints of humanoid fingertips. We introduce \textbf{GlowTact}, a pressure-responsive vision-based tactile sensing mechanism that directly visualizes contact pressure. \textbf{GlowTact} requires only single-color, non-directional illumination, and the raw tactile image directly represents the pressure distribution without explicit geometry reconstruction. This simple sensing principle enables compact, customizable tactile sensors while preserving high sensitivity and rich spatial detail. We demonstrate gram-scale contact detection, accurate normal-force estimation, and reconstruction of fine contact geometry, including M1 screw threads. These results establish \textbf{GlowTact} as a practical new sensing technology for compact humanoid fingertips, combining a durable nitrile membrane and simple optical design with sensitive and information-rich tactile perception.
Figures & tables
Fig. 1: GlowTact is a pressure-responsive vision-based tactile sensing mechanism . A GlowTact-equipped LEAP hand grasps a screw (left), while the tactile images (right) directly visualize contact pressure and fine thread geometry. GlowTact combines sensitive low-force response with rich spatial detail using only single-color illumination.
Fig. 2: Working principles of representative vision-based tactile sensors. (A) GelSight uses directional illumination and an opaque reflective membrane to encode surface geometry through shading. (B) 9DTact uses light attenuation in a translucent gel beneath a black membrane to encode local gel thickness in image intensity.
Fig. 3: GlowTact sensing principle. In the unloaded state (A) , a thin air gap separates the black membrane from the gel’s micro-textured surface, so light within the gel is diffusely reflected at the gel–air interface, and the camera sees a uniform light gray. When an indenter presses the membrane into optical contact with the gel (B) , the reflective interface is eliminated and light is instead absorbed by the black membrane, darkening the contact region as pressure increases.
Fig. 4: Exploded view of the Flat GlowTact sensor. A black membrane cut from a nitrile glove is stretched across the textured silicone gel pad, which is supported by the gel holder, forming the pressure-sensitive optical interface. A perimeter LED strip illuminates the interface through an acrylic plate, while a fisheye camera mounted in the camera holder observes it from within the sensor shell.
Fig. 5: Response of flat GlowTact sensor. Fingerprint images demonstrate high sensitivity and high resolution, shown for normal loads of 0.3 N, 1.0 N, and 2.0 N. Increasing force results in progressively stronger optical coupling and a larger contact area while preserving fine ridge patterns.
Fig. 6: Exploded view of the Omnidirectional GlowTact fingertip sensor. A black membrane cut from a nitrile glove is stretched across the surface of the textured silicone, which coats a clear epoxy skeleton. A single-color LED ring and fisheye camera sit at the base of the sensor.
Fig. 7: Humanoid GlowTact fingertip sensor. (a) Main components of the sensor: wide-angle camera, U-shaped LED strip, finger housing, and curved gel sensing body. (b) Assembled fingertip without and with the black nitrile contact membrane. (c) Raw tactile image produced by pressing a U.S. quarter against the sensor; dark regions indicate increased local pressure, while the coin edge and raised lettering remain visible.
Fig. 8: 3D reconstruction of GlowTact, compared against 9DTact [ 8 ] . Top row: representative test objects. GlowTact is evaluated using an M5 Phillips screw head, threaded screws (M3, M2.5, M2, M1.5, and M1), and a steel-ball array. Representative 9DTact results [ 8 ] are included for reference, with set screws (M4, M6, and M8) and a steel-ball array. Middle row: corresponding tactile images. Bottom row: reconstructed contact geometries. Despite relying on a fundamentally different sensing mechanism, GlowTact retains sufficient spatial information to recover the overall contact geometry.
Fig. 9: Comparison of low-load contact sensitivity between GlowTact (left) and GelSight Mini (right). Lightweight objects are gently placed on each sensor under their own weight without additional loading. For each object, the raw tactile image, the difference image, and a 3× contrast-enhanced difference image are shown. GlowTact produces clear contact responses for all tested objects, whereas GelSight Mini exhibits weaker image changes under the same passive loading conditions.
Fig. 10: Force sensitivity comparison between GlowTact and GelSight Mini. (a) Mean tactile response as a function of normal force. (b) Signal-to-noise ratio (SNR) versus force; the dashed line denotes the detection threshold ( SNR=3 ). (c) Minimum detectable force for each probe geometry. GlowTact produces stronger low-force responses, higher SNR, and consistently lower detection thresholds than GelSight Mini.
a Controlled-probe validation
Overall RMSE
Force-bin MAE
Sensor
0–0.5
0.5–2
2–20
GelSight Mini
0.285
0.067
0.088
0.185
GlowTact
0.276
0.056
0.058
0.209
TABLE I: Force-estimation errors at held-out spatial locations (N).
Department of Mechanical Engineering, National University of Singapore, Singapore · Department of Mechanical Engineering, Zhejiang University, Hangzhou, China