Point Cloud Sequence Encoding for Material-conditioned Graph Network Simulators

Philipp Dahlinger, Balázs Gyenes, Niklas Freymuth, Tobias Würth, Tai Hoang, Johannes Mitsch, Luise Kärger, Gerhard Neumann

Keywords: Pointcloud Sequence encoding, Graph Network Simulation, Material Estimation, Transformer, GNN

TL;DR: PEACH (Point cloud Encoding for Accurate Context Handling) predicts physical properties from point clouds to enable accurate neural simulation of unseen materials.

Abstract:

Graph Network Simulators (GNS) have emerged as powerful surrogates for complex physics-based simulation, offering inherent differentiability and orders-of-magnitude speedups over traditional solvers.
However, GNS typically assume access to the underlying material parameters, such as stiffness or viscosity, and rely heavily on dense mesh-based data.
These constraints severely limit their utility in realistic experimental settings.
While recent meta-learning approaches address the parameter dependency by inferring properties from mesh trajectories, they still require mesh-based trajectories as contexts.
In this work, we introduce Point cloud Encoding for Accurate Context Handling (PEACH), a novel framework that decouples parameter estimation from temporal dynamics.
PEACH infers latent physical properties directly from sequences of point clouds using a spatial-temporal encoder.
This parameter estimate then conditions a trajectory-level GNS, enabling accurate in-context simulation of unseen materials during inference.
Experiments on simulated object deformation tasks demonstrate that PEACH achieves simulation accuracy matching purely mesh-based baselines, while relying on more realistic and accessible data.

Journal Opt In: No, I do not wish to participate

Submission Number: 8