Contact-Centric Deformation Learning

[Project website] [Video] [Data]

Abstract

We propose a novel method to machine-learn highly detailed, nonlinear contact deformations for real-time dynamic simulation. We depart from previous deformation-learning strategies, and model contact deformations in a contact-centric manner. This strategy shows excellent generalization with respect to the object's configuration space, and it allows for simple and accurate learning. We complement the contact-centric learning strategy with two additional key ingredients: learning a continuous vector field of contact deformations, instead of a discrete approximation; and sparsifying the mapping between the contact configuration and contact deformations. These two ingredients further contribute to the accuracy, efficiency, and generalization of the method. We integrate our learning-based contact deformation model with subspace dynamics, showing real-time dynamic simulations with fine contact deformation detail.

Requirements: python3, h5py numpy

Meshes

meshes.h5

Key	Key	Description	Dimension
jelly	W	Linear basis (BGBC)	[num_jelly_nodes, num_handle_rows]
	q_ref	Reference transformations	[num_handle_rows, dim]
	tris	Triangle indices	[num_jelly_tris, dim]
	tets	Tetrahedron indices	[num_jelly_tets, dim+1]
star	W	Linear basis (rigid)	[num_star_nodes, dim+1]
	z_ref	Reference transformation	[dim+1, dim]
	tris	Triangle indices	[num_star_tris, dim]

Dataset

dataset.h5

Key	Description	Dimension
q	Handle transformations	[num_samples, num_handle_rows, dim]
z	Collider transformation	[num_samples, dim+1, dim]
x	Simulated node positions	[num_samples, num_jelly_nodes, dim]

Citation

@article {romero2022contactcentriclearning,
    author  = {Romero, Cristian and Casas, Dan and Chiaramonte, Maurizio M. and Otaduy, Miguel A.},
    title   = {{Contact-Centric Deformation Learning}},
    number  = "4",
    volume  = "41",
    journal = {ACM Transactions on Graphics (Proc. of ACM SIGGRAPH)},
    year    = {2022}
}