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.

Simulation

Method 1: Compiling the simulator

The installation of simulation can refer to cloth-funnels.

Method 2: Use the Docker image we provide.

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 

Model

The model checkpoint is shared via Baidu Netdisk Access code: nwii

Evaluate

 . ./eval_ppo.sh clothppo

Train

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 

Acknowledgements

• 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},
    }