ChirikjianLab/deformable_ravens_vf

Goal-Conditioned Transporter Networks (GCTN) for the multi-modal action proposal module of Transporters with Visual Foresight (TVF)

NOTE: this code is based heavily on the Ravens code base from Google and retains the same license.

NOTE: this code is based heavily on the orignal DeformableRavens repo.

DeformableRavens-VF

Code for training the Goal-Conditioned Transporter Network (GCTN) of the multi-modal action proposal of Transporters with Visual Foresight (TVF). The main repository for TVF is ravens_visual_foresight. It also contains the data and pretrained models. This code has been tested on Ubuntu 20.04 and Python 3.8. If you have any questions, please use the issue tracker.

TVF repo || Paper Link || Project Website

Installation

./install_python_ubuntu.sh

Usage

To train the GCTN for multiple tasks

python train_multi_task.py --data_dir=ravens_visual_foresight/data_train --models_dir=ravens_visual_foresight/gctn_models --num_demos=10 --num_runs=1

--data_dir specifies the directory of the training data. --models_dir specifies the directory to save the trained models. --num_demos specifies the number of demos per training task used for training. --num_runs specifies the number of training runs.

Related Work

• Learning to Rearrange Deformable Cables, Fabrics, and Bags with Goal-Conditioned Transporter Networks

• Transporter Networks: Rearranging the Visual World for Robotic Manipulation