FrankaAllegro-Policies

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 .

1. Collect Demos using OpenTeach.
2. Change the path in configs to the path where you saved your data.
3. Preprocess the data using the following command python3 preprocess.py
4. Choose the path you have collected data in configs in data_dir:. Choose the data representations you want to use for training from image/tactile,allegro,franka
5. Once preprocessed you can train the Vision and tactile encoder using python train.py. You can edit train.yaml accordingly with the choice of encoder, rl_learners, rewarders and optimizers.
6. After training the Vision and tactile encoders you can start the offset learning following TAVI using python train_online.py.
7. You can set the task, base_policy,agent, explorer and rewarder. configs

Citation

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}
}