GelSight: Frequently Asked Questions

What’s the idea?

Coat a slab of gel elastomer with stretchy reflective skin. Press an object against it. The surface distorts, and you can view it from behind. In effect, we are coating the object with a known BRDF.

Why is this useful?

You can capture shape and texture in real time, using a simple, compact system.

What’s the advantage?

The sensor is simple, cheap, and fast. It gets extremely fine surface detail. It has no problem with transparent or shiny objects, which can confuse laser rangefinders.

What’s the disadvantage?

It works best for small objects without too much depth. It’s not a general purpose 3D scanner.

What’s the gel made of?

You can use silicone, but we usually use a thermoplastic elastomer (TPE), like the stuff in gel insoles or padded bike seats. For example, you can use Ultraflex from, and melt in into a shape.

How do you make the reflective “skin”?

Typically we dissolve the elastomer in solvent, add pigment to make a paint, and spray it on. Metal flake pigment yields a specular paint that has high contrast for fine details.

Can it be big? Small? Hard? Soft?

Yes, yes, yes, yes.

How do you measure 3D?

The GelSight sensor turns pressure into shape, and turns shape into an image. We get 3D using machine vision, specifically “photometric stereo.” Using 3 colored lights, an RGB camera, and a calibration target, we build a lookup table that maps color (R,G,B) to gradient. We reconstruct the surface from its gradients by solving Poisson’s equation.

What is the resolution limit?

Not sure, but currently we can go below 10 microns. (A human hair is 50-100 microns). See the picture of oregano below.

What other applications?

Fingerprint reader, multitouch input device, tactile sensor for robotics, medical devices, etc.

How did you think of this?

It’s a long story.