This repository contains the policy learning code for the Franka-Allegro Robot environment with TAVI.
This codebase has been released as a part of OpenTeach
Clone the repository using the following command.
git clone https://github.com/NYU-robot-learning/FrankaAllegro-Policies.git
Installation
mamba env create -f conda_env.yml
Install the Codebase as a module using
pip install -e .
- Collect Demos using OpenTeach.
- Change the path in configs to the path where you saved your data.
- Preprocess the data using the following command
python3 preprocess.py
- Choose the path you have collected data in configs in
data_dir:
. Choose the data representations you want to use for training fromimage/tactile,allegro,franka
- Once preprocessed you can train the Vision and tactile encoder using
python train.py
. You can edittrain.yaml
accordingly with the choice of encoder, rl_learners, rewarders and optimizers. - After training the Vision and tactile encoders you can start the offset learning following TAVI using
python train_online.py
. - You can set the task, base_policy,agent, explorer and rewarder. configs
If you use this repo in your research, please consider citing the paper as follows:
title={OPEN TEACH: A Versatile Teleoperation System for Robotic Manipulation},
author={Aadhithya Iyer and Zhuoran Peng and Yinlong Dai and Irmak Guzey and Siddhant Haldar and Soumith Chintala and Lerrel Pinto},
year={2024},
eprint={2403.07870},
archivePrefix={arXiv},
primaryClass={cs.RO}
}