The Pulse

MIT reports WorldParticle model for simulating varied materials

MIT News describes WorldParticle, a system that uses one AI architecture to simulate materials including fluids, granular substances and proteins. The researchers report tests across six physical phenomena and say protein structures were mo

AI.info Team · Written with AI · fact-checked · 2026-11-10

One system for different materials

Researchers at MIT, the University of British Columbia and other institutions have developed an AI system designed to simulate a range of physical processes using a single architecture, MIT News reported on Nov. 9. Called WorldParticle, the system represents materials as particles and predicts how they move from one simulation step to the next.

The researchers say the approach could model processes that normally call for separate tools, including sliding sand, folding proteins and fluid moving against a deformable solid. They also describe possible uses in studying how sloshing oil might affect an elastic storage-tank valve, or how sand or soil might move when touched by a soft robot.

Particles in, predicted motion out

WorldParticle represents an object as thousands or millions of tiny, three-dimensional particles. It tracks particle position and velocity over space and time; a neural network takes particle positions and physical properties as input and predicts their state for the next frame.

Conventional particle-based simulations typically start with equations describing how a particular material behaves. Researchers then convert those equations into forms that numerical solvers can process, repeating the work for each material and process. The MIT team instead designed a neural network to learn how to solve governing equations in general, the report says.

The system separates external forces from the simulated process before making a prediction, then adds those forces back in to adjust particle movement. According to the researchers, this setup helps it generalize to materials, forces and environmental conditions that were not present in training.

Peter Yichen Chen, a former MIT postdoctoral researcher and now an assistant professor at UBC, compared the architecture with ChatGPT. Rather than predicting the next word in a text sequence, he said, WorldParticle predicts where particles will move, one step at a time.

Tests and limits

The team demonstrated the system on six physical phenomena, including deformable objects, granular materials and proteins. MIT News reported that the researchers found higher accuracy than numerical simulation methods in those tests.

In one granular-material example, WorldParticle modeled a material collapsing into a pile, while some baseline methods left sand particles floating in the air. Co-corresponding author Minghao Guo said the system modeled changes in protein structures about 100 times faster than numerical simulators.

The researchers say they want to extend it to areas including soft-robot manipulation, RNA evolution and astrophysical phenomena such as merging galaxies.

Next steps

Guo and his colleagues also want to build toward a general-purpose simulation pipeline for scientific discovery. The research is due to be presented at the ACM SIGGRAPH Conference and Exhibition on Computer Graphics and Interactive Techniques in Asia.

Guo is an MIT postdoctoral researcher and a co-corresponding and co-lead author. The paper, co-authored by researchers from MIT, UBC and other institutions, is titled “WorldParticle: Unified World Simulation of Lagrangian Particle Dynamics via Transformer.”