The contact layer · vision-based touch

A tactile fingertip reconstructs what it touches

An elastomer gel deforms against an object; a camera under three colored lights sees only color. Geometry — depth, force, shear — is computed, by photometric stereo, on the device. Press a shape in; drag the raw pane to shear.

Contact area
Normal force
Shear
Grip state
Recon fidelity
◈ Recognized as
1 · Raw sensor — gel under R/G/B lights
What the camera actually sees. Surface slope is encoded in color. Drag to apply shear ↔.
2 · Recovered normals — photometric stereo
n = M⁻¹·[R,G,B]. Per-pixel surface orientation, solved from the three lights.
3 · Reconstructed depth — ∇²z = ∇·g
Normals integrated to a height map (Poisson). The micro-geometry, from touch alone.
4 · Grip, slip & haptic out
Markers flow with the gel; the gold ring is the stuck core, shrinking as slip nears. Red = slipping. The trace is the vibration a haptic actuator would replay.
A GelSight/DIGIT-class fingertip: an image sensor becomes a geometry-and-force sensor. Meta's DIGIT 360 packs ~8.3M taxels into one hemisphere.