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

Taxim is an example-based simulator for GelSight tactile sensors and its variations. For more information of Taxim, you can check the paper or the webpage.

Installation and Prerequisites

Basic dependencies: numpy, scipy, matplotlib, cv2

To install dependencies: pip install -r requirements.txt

Optional dependencies: ros with usb-cam driver (to collect the tactile images from a tactile sensor), nanogui (to annotate the raw data.)

To install ros usb-cam driver, please check out here.

To install nanogui, please check out here.

Usage

If you want to customize the Taxim on your own sensor, please follow the DataCollection and Calibration to calibrate the Taxim and generate calibration files. And modify the parameters under Basic.params and Basic.sensorParams accordingly.

We provide a set of calibration files and you can work with them directly. You can follow instruction of Optical Simulation and Marker Motion Field Simulation to start working with the provided examples. And feel free to change the parameters under Basic.params.

Data Collection (optional)

Connect a GelSight sensor with your pc and launch the camera driver.
Change the self.gel_sub in gelsight.py to your sensor camera's topic.
Run python record_Gel.py and input the file name and number of frames to collect the data.

Calibration (optional)

Generate data pack: Run python generateDataPack.py -data_path DATA_PATH where DATA_PATH is the path to the collected raw tactile data. Hand annotate the contact center and radius for each tactile image. dataPack.npz will be saved under the DATA_PATH.
Generate polynomial table: Run python polyTableCalib.py -data_path DATA_PATH where DATA_PATH is the path to the data pack. polycalib.npz will be saved under the DATA_PATH.
Generate shadow table: Run python generateShadowMasks.py -data_path DATA_PATH where DATA_PATH is the path to the collected shadow calibration images. shadowTable.npz will be saved under the DATA_PATH.
Generate FEM tensor maps: Export the ANSYS FEM displacement txt files and set the path in the main function. Run python generateTensorMap.py and femCalib.npz will be saved under calibs folder.

All the calibration files from a GelSight sensor have been provided under calibs folder.

Optical Simulation

You can input a point cloud of a certain object model and define the pressing depth, or directly input a depth map. All the parameters in Basic.params are adjustable. depth is in millimeter unit.

Run python simOptical.py -obj square -depth 1.0 to visualize the examples. Results are saved under results.

Marker Motion Field Simulation

You can input a point cloud of a certain object model and define the loads on x, y, z directions. dx and dy are shear loads and dz is normal loads, which are all in millimeter unit. Run python simMarkMotionField.py -obj square -dx 0.3 -dy 0.4 -dz 0.5 to visualize the resultant displacements. Results are saved under results.

Operating System

Taxim has been tested on macOS Catalina (10.15.7) and Ubuntu (18.04.1) with anaconda3.

Configuration for MacOS: python 3.8.5, numpy 1.20.1, scipy 1.6.1, opencv-python 4.5.3.56

Configuration for Ubuntu: python 3.6.13, numpy 1.19.5, scipy 1.5.4, opencv-python 4.5.2.54

License

Taxim is licensed under MIT license.

Citating Taxim

If you use Taxim in your research, please cite:

@article{si2021taxim,
  title={Taxim: An Example-based Simulation Model for GelSight Tactile Sensors},
  author={Si, Zilin and Yuan, Wenzhen},
  journal={arXiv preprint arXiv:2109.04027},
  year={2021}
}

xEnVrE/Taxim