The reflective membrane on the gel surface deforms to the shape of the touched object. A camera under the gel captures the membrane lit from several directions by multi-color LEDs; from the per-channel brightness the system computes local surface slopes (photometric stereo) and, by integration, a height map (the 3D geometry of the imprint). Motion of the printed markers is used to estimate shear forces and slip. The results feed grasp-control algorithms or manipulation-learning networks.
Robots need dense, accurate tactile information for dexterous manipulation that taxel arrays cannot provide. GelSight delivers a micrometer-scale map of contact geometry and forces from a single, low-cost camera module.
A transparent elastomer coated with an opaque, reflective layer.
Official
Capture the deformed membrane from different directions/colors.
Official
Recover 3D geometry and shear forces and slip.
Official
The reflective membrane wears out and needs replacement.
Accuracy depends on correct lighting and optics calibration.
Adelson's lab demonstrates a high-resolution gel-based vision tactile sensor.
GelSight increasingly used for manipulation and in-hand recognition.
A compact commercial module popularizes the sensor in research and robotics.
