This new approach, discussed at SIGGRAPH, studies a contact-centric manner.
Check out a paper prepared for this year's SIGGRAPH that discusses a novel method to machine-learn highly detailed, nonlinear contact deformations for real-time dynamic simulation. This new project proposes a way to 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," wrote the research team. "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."
The team states that these two ingredients allowed them to boost the accuracy, efficiency, and generalization of the method. One of the demos, for example, shows how the team used the MANO kinematic model that interacts with a 3D cube and generates fine deformations depending on interactions.
You can find the project page here. What are your thoughts on the approach? Are there some other parameters that should be added to the model?
Don't forget to join our Reddit page and our Telegram channel, follow us on Instagram and Twitter, where we are sharing breakdowns, the latest news, awesome artworks, and more.