Physics-Encoded Graph Neural Networks for
Deformation Prediction under Contact

Mahdi Saleh1 Michael Sommersperger1 Nassir Navab1 Federico Tombari1,2
1Technical University of Munich 2Google
ICRA 2024
Preprint Dataset Code
Teaser Image

We focus on predicting deformation during contact between a rigid mesh and a soft mesh, based on their physics properties. We transform these inputs into physics-Encoded graphs, and utilize Graph Neural Networks to calculate the new positions of the soft body after deformation.

Architecture Overview

Our approach transforms rigid and soft meshes into graphs with nodes representing key features. These graphs are processed by a GNN encoder, followed by cross-attention-based feature interaction. The final step is decoding these features to predict new mesh positions post-contact.

Abstract

In the field of robotics, it's important to understand object deformation during tactile interactions. Precise understanding of deformation can elevate robotic simulations and have broad implications across different industries. We introduce a method using Physics-Encoded Graph Neural Networks (GNNs) for such predictions. Similar to robotic grasping and manipulation scenarios, we focus on modeling the dynamics between a rigid mesh contacting a deformable mesh under external forces. Our approach represents both the soft body and the rigid body within graph structures, where nodes hold physical states of the meshes. We also incorporate cross-attention mechanisms to captures the interplay between the objects. By jointly learning the geometry and physics, our model reconstructs consistent and detailed deformations. We release the code alongside the dataset designed to analyze the geometry and physical properties of deformable everyday objects, improving robotic simulation and grasping research.

Short Video Presentation

BibTeX

@article{saleh2024physics,
        title={Physics-Encoded Graph Neural Networks for Deformation Prediction under Contact},
        author={Saleh, Mahdi and Sommersperger, Michael and Navab, Nassir and Tombari, Federico},
        journal={arXiv preprint arXiv:2402.03466},
        year={2024}
      }