Taxim: An Example-based Simulation Model for GelSight Tactile Sensors

An example-based simulation model for GelSight visuotactile sensors, calibrated from a small amount of real data.

Taxim was proposed in 2021 by Zilin Si and Wenzhen Yuan at Carnegie Mellon University's RoboTouch lab, built specifically to simulate GelSight-type visuotactile sensors. “Example-based” means it does not model the optical path from scratch; instead it calibrates a polynomial lookup table from real samples captured on the actual sensor, mapping gel-pad deformation geometry directly to camera pixel brightness, while the motion of markers printed on the gel is computed separately by superimposing elastic-deformation theory. Calibration needs fewer than 100 real data points, so it is easy to port to different GelSight models. The paper reports lower per-pixel intensity error than prior methods, and it runs on a CPU. It is often compared with TACTO: TACTO renders images with a general-purpose renderer, while Taxim calibrates its output from measured data.

ExampleGive Taxim an object's point cloud and indentation depth, and it returns the corresponding GelSight tactile image; given loads along the x, y, and z directions, it can also return the resulting displacement field of the gel's surface markers.

Also called

Taxim Simulation Model

Related

Tactile Simulation · GelSight · TACTO: A Fast, Flexible, and Open-source Simulator for High-Resolution Vision-based Tactile Sensors · Marker Tracking · Photometric Stereo

Sources

Taxim: An Example-based Simulation Model for GelSight Tactile Sensors (arXiv 2109.04027)
Robo-Touch/Taxim GitHub 仓库 (Chinese)

As of

2021-12