TACTO: A Fast, Flexible, and Open-source Simulator for High-Resolution Vision-based Tactile Sensors

Meta's open-source visuotactile simulator, using PyBullet to render tactile images for sensors like DIGIT.

TACTO was developed by Shaoxiong Wang, Mike Lambeta, Po-Wei Chou, and Roberto Calandra, with code released in Meta's facebookresearch repository; the paper appeared in IEEE RA-L and was presented at ICRA 2022. It uses the PyRender renderer to generate hundreds of high-resolution tactile images per second from the contact geometry between an object and an elastic gel pad, and provides an interface that connects to the PyBullet physics engine, shipping with built-in models and configurations for the DIGIT and OmniTact sensors. The authors note explicitly that TACTO does not itself model the physical accuracy of contact dynamics (deformation, friction) — that is left to existing physics engines. It is open source under the MIT license and installs directly with pip.

ExampleThe paper used TACTO to simulate 1 million grasps for training a grasp-stability predictor, and also demonstrated a task using touch to control a rolling marble, along with preliminary sim-to-real transfer.

Also called

TACTO Simulator

Related

Tactile Simulation · DIGIT · PyBullet · Taxim: An Example-based Simulation Model for GelSight Tactile Sensors · Vision-Based Tactile Sensor

Sources

TACTO (arXiv 2012.08456)
facebookresearch/tacto GitHub 仓库 (Chinese)

As of

2022-02