Physics-Encoded Graph Neural Networks for Deformation Prediction under Contact

This repository contains the implementation of a model that predicts deformation of shapes using physics encoded graph neural networks. Our approach has been accepted at ICRA 2024, and you can find the preprint of our paper on arXiv.

Prerequisites

• Anaconda or Miniconda: Ensure Anaconda or Miniconda is installed on your system. Download it here.
• Weights & Biases Account: Needed for experiment tracking. Sign up here.

Installation and Setup

1. Clone the Repository:

   git clone https://github.com/mahdi-slh/DeformContact.git
   cd DeformContact

Replace username with your GitHub username and repository with the name of your repository.

2. Run Setup Script:

    bash setup.sh
   

This script will create and activate a conda environment named deform, and install the necessary packages.

3. Setup Weights & Biases: Login to your Weights & Biases account:

    wandb login

Follow the on-screen instructions.

4. Download the dataset Please download the dataset from here and place it in the following directory within the cloned repository.

    python visualize.py

Ensure the config_path variable in the main function of visualize.py is set to the path of your config file.

5. Visualize the data Run the visualization script:

    python visualize.py

Ensure the config_path variable in the main function of visualize.py is set to the path of your config file.

6. Train the Model Start model training:

    python train.py

Ensure the config_path variable in the main function of train.py is set to the path of your config file.

7. Evaluate the Model Evaluate the trained model:

    python eval.py