Skip to content

s-tian/bench-press

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

This repository contains code for operating the tactile sensor testbench (modified 3 axis CNC), as well as model training code written in PyTorch.

It accompanies the paper OmniTact: A Multi-Directional High Resolution Touch Sensor, by Akhil Padmanabha, Frederik Ebert, Stephen Tian, Roberto Calandra, Chelsea Finn, and Sergey Levine, to appear in the International Conference on Robotics and Automation (ICRA) 2020.

Paper website: https://sites.google.com/berkeley.edu/omnitact

Instructions

To run a control experiment on a dummy environment or on the testbench, create a configuration file and run it per the following example:

python run/run.py <experiment_config> <num_trajectories>

For a concrete example, the command python run/run.py keyboard_control_dummy 5 rolls out 5 trajectories in a dummy environment, using a policy which queries the user to enter actions via the keyboard.

To train a model using data collected using the run script, create a model configuration file and train the model using

python scripts/train_policy.py <model_config>

for example

python scripts/train_policy.py models/experiments/y_reg/y_reg_rand.yaml

trains a model which regresses the XYZ location of the end effector from tactile images.

Further hardware, electronics, firmware, and software documentation is provided in /doc/README.md.

About

OmniTact: A Multi-Directional High Resolution Touch Sensor (ICRA 2020)

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published