Simulating physical reality has long been a fragmented affair. A scientist modeling ocean waves needs one solver; a computational biologist watching a protein fold needs another; an engineer studying sand dunes needs a third. Each tool is painstakingly hand-coded around the specific equations that govern one narrow phenomenon.
Researchers at MIT, the University of British Columbia and a slate of collaborating institutions have now unveiled a system designed to replace that patchwork with a single architecture. The system, called WorldParticle, uses one unified deep-learning model to simulate widely different physical processes — from sliding sand to folding proteins — according to MIT News. For more context on this story, see our ongoing AI trends.
The work, an open-access paper titled "WorldParticle: Unified World Simulation of Lagrangian Particle Dynamics via Transformer," will be presented at the ACM SIGGRAPH Conference and Exhibition on Computer Graphics and Interactive Techniques in Asia. The collaboration spans MIT, UBC, Georgia Tech, Inria, Meta, the University of Pennsylvania, the University of Utah, UCLA and the University of Hong Kong.
A 'GPT Moment' for Simulation
"We want to have a GPT moment for simulation," said Peter Yichen Chen, a former MIT postdoc who is now an assistant professor at UBC and a co-corresponding author of the paper, in MIT News' report. "Every time people want to build a simulator for a particular physical phenomenon, they need to code it up from scratch. Our vision is to have a foundation model that works out of the box for any scenario you want to simulate. This is a first step in that direction."
The reference to GPT is not accidental. WorldParticle's architecture borrows the same sequential-prediction idea that powers large language models. "It is a similar architecture to ChatGPT. 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 joined by co-corresponding and co-lead author Minghao Guo, an MIT postdoc, and co-lead authors Caoliwen Wang and Siyuan Chen, graduate students at UBC. The team also includes Wojciech Matusik, an MIT professor of electrical engineering and computer science who leads the Computational Design and Fabrication Group within the Computer Science and Artificial Intelligence Laboratory (CSAIL).
How WorldParticle Works
The system rests on a long-standing idea in computational physics: the Lagrangian representation, in which a material is described as a collection of discrete particles whose positions and velocities evolve over time. WorldParticle breaks each object it simulates into thousands or millions of tiny three-dimensional dots, then tracks how those particles move through space and time.
Learning the Equations, Not Hand-Coding Them
The departure from tradition lies in what writes the rules. In a classical pipeline, a scientist starts with a governing equation that defines how a material behaves, then converts that complex mathematics into a formula a numerical solver can handle — repeating the entire exercise for every new material or process.
WorldParticle instead uses a neural network that learns to solve governing equations in general. The network takes the 3D positions and physical properties of the particles as inputs and outputs their positions and properties for the next frame of the simulation, predicting the material's motion over time and space in one-second increments.
Handling External Forces
Real-world materials are constantly pushed around by forces a model cannot fully anticipate — an ocean wave slamming into a seawall, or a foot pressing into sand. Before each simulation step, WorldParticle separates out the external forces acting on the system. After predicting the particles' initial states, the system integrates those forces back into the simulation, correcting its predictions based on how the forces affect the object's behavior.
That correction loop, the researchers report, is what allows the model to generalize to materials, environmental conditions and forces it never encountered during training.
Compressing Millions of Particles
Scale posed the central engineering challenge. A complex object can shatter into millions of particles, quickly overwhelming the model's computational budget. To keep processing tractable, the team employed a merging technique that combines and compresses particles while preserving fine-grained detail in their behavior.
"We needed to do a lot of detailed engineering and employ many tweaks to the architecture for it to be truly generalizable and scalable," Guo told MIT News.
Tested Against Six Physical Phenomena
The researchers demonstrated WorldParticle on six different physical phenomena, including deformable objects, granular materials and proteins, reporting higher accuracy than established numerical simulation methods.
The examples were concrete. In one granular test, WorldParticle captured how a pile of material collapses under gravity while some baseline methods left sand particles floating improbably in the air. On molecular dynamics, the system modeled how protein structures change over time roughly 100 times faster than numerical simulators, according to Guo.
The practical payoff runs in both directions: faster results for problems scientists already study, and the ability to couple phenomena that previously demanded separate simulators. The team highlights scenarios such as fluid sloshing against a deformable solid — for instance, how sloshing oil might stress an elastic valve inside a storage tank — where the fluid and the structure it pushes against must be modeled together.
From Soft Robots to Merging Galaxies
The framework's generality points to applications well beyond computer graphics. The researchers say the same approach could inform soft robot manipulation tasks, in which a compliant machine must predict how sand or soil will shift under its touch. They also list RNA evolution and even astrophysical phenomena such as the merging of galaxies as candidate domains.
In the longer term, the team hopes the work helps scientists discover new materials and explore physical processes that existing simulation methods cannot effectively study.
"We see the unification of computation and science giving us new insights and tools to discover physical phenomena or explain some unexplainable physical phenomena," Guo said. "With a unified model, we could unlock many discoveries that we can't even imagine."
A First Step, Not a Finished Foundation
The researchers are careful to frame WorldParticle as an opening move rather than a completed foundation model. Moving forward, they plan to adapt the system to more physical systems and build toward a full, general-purpose simulation pipeline for scientific discovery.
If that arc sounds familiar, it is by design. Language models spent years as narrow, per-task systems before scaling into general-purpose tools — and the WorldParticle team is explicitly betting that physical simulation can follow a similar path, one particle step at a time.
This article is based on reporting by Adam Zewe, published by MIT News on November 9, 2026.---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
