Tactile Image

Arranging tactile sensor readings into a 2D image, so ordinary vision models can process them directly.

A tactile image organizes a tactile signal into 2D image form. It comes from two sources: a tactile array, where each element’s reading becomes one pixel, giving a low-resolution pressure-distribution map; or a vision-based tactile sensor such as GelSight or DIGIT, whose internal camera directly photographs the soft gel deforming under pressure, which is already a color image that clearly shows the texture and shape of whatever it touched. The benefit of framing it as an image is that off-the-shelf vision models — convolutional networks, ViTs — can be applied directly, and it can be fed into a policy alongside camera images for combined vision-touch fusion. A sequence of tactile images sampled over time can also capture how a contact evolves, useful for detecting slip or estimating shear force.

ExamplePressing a GelSight onto a coin produces a tactile image that clearly shows the coin’s embossed relief, from which photometric stereo recovers the local 3D shape.

Also called

Pressure Distribution Map

Related

Tactile Array · Vision-Based Tactile Sensor · GelSight · Visuo-Tactile Fusion · Tactile Representation Learning · Photometric Stereo

Sources

GelSight: High-Resolution Robot Tactile Sensors for Estimating Geometry and Force (Yuan et al., Sensors 2017)