Visuo-Skin (ViSk)

Accompanying code for training VisuoSkin policies as described in the paper: Learning Precise, Contact-Rich Manipulation through Uncalibrated Tactile Skins

About

ViSk is a framework for learning visuotactile policies for fine-grained, contact-rich manipulation tasks. ViSk uses a transformer-based architecture in conjunction with AnySkin and presents a significant improvement over vision-only policies as well as visuotactile policies that use high-dimensional tactile sensors like DIGIT.

Installation

1. Clone this repository

git clone https://github.com/raunaqbhirangi/visuoskin.git

2. Create a conda environment and install dependencies

conda create -f env.yml
pip install -r requirements.txt

3. Move raw data to your desired location and set DATA_DIR in utils.py to point to this location. Similarly, set root_dir in cfgs/local_config.yaml.

4. Process data for the current-task (name of the directory containing demonstration data for the current task) and convert to pkl.

python process_data.py -t current-task
python convert_to_pkl.py -t current-task

5. Install xarm-env using pip install -e envs/xarm-env

6. Run BC training

python train_bc.py 'suite.task.tasks=[current-task]'