ClothPPO: A Proximal Policy Optimization Enhancing Framework for Robotic Cloth Manipulation with Observation-Aligned Action Spaces.
International Joint Conference on Artificial Intelligence 2024
Project Page | Video | Arxiv
This repository contains code for training and evaluating in simulation for Ubuntu 20.04. It has been tested on machines with Nvidia GeForce RTX 4090.
The installation of simulation can refer to cloth-funnels.
If you are familiar with Docker, we have also open-sourced the Docker image for ClothPPO. You can use ClothPPO through Docker without the need to compile the simulation or configure the conda environment yourself.
docker pull elcarimqaq/clothppo
The model checkpoint is shared via Baidu Netdisk Access code: nwii
. ./eval_ppo.sh clothppo
ClothPPO was trained on a single RTX 4090 GPU. You can modify the training script we provide, train_ppo.sh
, as needed.
. ./train_ppo.sh
- This codebase is heavily built on on cloth-funnels.
- The cloth simulator is a fork of PyFlex from Softgym
If you find this codebase useful, consider citing:
@inproceedings{ijcai2024p762,
title = {ClothPPO: A Proximal Policy Optimization Enhancing Framework for Robotic Cloth Manipulation with Observation-Aligned Action Spaces},
author = {Yang, Libing and Li, Yang and Chen, Long},
booktitle = {Proceedings of the Thirty-Third International Joint Conference on
Artificial Intelligence, {IJCAI-24}},
publisher = {International Joint Conferences on Artificial Intelligence Organization},
editor = {Kate Larson},
pages = {6895--6903},
year = {2024},
month = {8},
note = {Main Track},
doi = {10.24963/ijcai.2024/762},
url = {https://doi.org/10.24963/ijcai.2024/762},
}