MIT and UBC Researchers Build WorldParticle, One AI Model That Simulates Sand, Fluids and Folding Proteins

Researchers at the Massachusetts Institute of Technology and the University of British Columbia built one artificial intelligence system that simulates sand, fluids and proteins, MIT said Monday, Nov. 9. The deep-learning system is called WorldParticle. It will be presented at the ACM SIGGRAPH conference in Asia, according to MIT.

Related: MIT Team Uses Floppy Connectors to Build Nanoparticle Crystals Never Seen Before.

What the system does

Scientists today usually pick a separate tool for each physical process, such as one solver for flowing water and another for protein folding, MIT said. Each tool is hand-coded around the equations for that process, according to MIT.

WorldParticle replaces that approach with one architecture, the university said. It represents every material as tiny particles and predicts where those particles will move, one step at a time.

“It is a similar architecture to ChatGPT,” said Peter Yichen Chen, a co-corresponding author and assistant professor at the University of British Columbia. “But instead of predicting one word at a time to write a piece of text, we are predicting how these particles will move, one step at a time,” Chen said. Chen is a former MIT postdoc.

How it learns

Scientists who model particles typically rely on a governing equation for each material, then convert it into a form a numerical solver can handle, MIT said. The team instead designed a neural network that learns to solve governing equations in general, according to the university.

The network takes the 3D positions and physical properties of particles as inputs and outputs their next positions and properties, MIT said. The system separates external forces before each step and adds them back afterward, according to the release. MIT said that design lets it generalize to materials and forces it did not encounter in training.

A complex object can break into millions of particles, which makes the computation heavy, MIT said. The researchers used a merging technique that combines and compresses particles while keeping fine detail, according to the university.

What the tests showed

The team modeled six physical phenomena, including deformable objects, granular materials and proteins, with higher accuracy than numerical simulation methods, MIT said. In one test, WorldParticle captured how a granular object collapses into a pile, while some baseline methods left sand particles floating in the air, according to the release.

The method modeled how protein structures change over time about 100 times faster than numerical simulators, said Minghao Guo, an MIT postdoc and co-lead author.

These results are the researchers’ own account, as described in the MIT release.

What could come next

The system could help engineers model how sloshing oil affects an elastic valve in a storage tank, MIT said. It could also help them study how sand or soil moves when a soft robot touches it, according to the university.

The researchers said the framework could apply to soft robot manipulation, RNA evolution and astrophysical events such as galaxy mergers. They want to build toward a full general-purpose simulation pipeline for scientific discovery, MIT said.

“We want to have a GPT moment for simulation,” Chen said. He called the work a first step toward a foundation model that works for any scenario.

Who was involved

The authors include Guo, Chen and University of British Columbia graduate students Caoliwen Wang and Siyuan Chen, MIT said. Wojciech Matusik, an MIT professor who leads the Computational Design and Fabrication Group at the Computer Science and Artificial Intelligence Laboratory, is also an author. Other authors are at Georgia Tech, Inria, Meta, the University of Pennsylvania, the University of Utah, the University of California at Los Angeles and the University of Hong Kong.

For more coverage, see AI News, Technology and Science.

Sources

• MIT News: A single AI model simulates the behavior of many substances, from sliding sand to folding proteins

About the reporter: Theo Lin is an AI reporter covering AI, technology and science for The Boston Report. This article was produced with AI under our editorial policy. Read our AI and media disclosure.