Researchers at MIT, the University of British Columbia and other institutions have developed WorldParticle, an AI model that simulates several kinds of physical behavior using one system. MIT says it could help scientists and engineers study materials that typically require separate simulators, from shifting sand to changing protein structures.
WorldParticle represents a material as thousands or millions of three-dimensional particles. A neural network uses their positions and physical properties to predict where they will be in the next step of a simulation. It then accounts for external forces, such as an impact, when updating its predictions.
Conventional simulators are built around equations for a particular material or process. According to MIT, WorldParticle learns how particles move instead of requiring researchers to build a new model and solver for each case. The team also developed a way to merge and compress particles so the system can handle complex objects.
Tests showed the model was faster and more accurate
The researchers tested WorldParticle on six types of physical phenomena, including deformable objects, granular materials and proteins. MIT says it produced more accurate results than the numerical simulation methods used for comparison. For protein structures, MIT postdoc Minghao Guo says it modeled changes over time about 100 times faster than numerical simulators.
The team says a shared simulation system could help engineers study interactions between different materials, such as fluid pressing against a flexible valve or a soft robot touching soil. Those are potential uses, rather than results reported from the six tests.
The researchers plan to adapt WorldParticle to more physical systems and work toward a general-purpose simulation pipeline. Their open-access paper will be presented at the ACM SIGGRAPH Conference and Exhibition on Computer Graphics and Interactive Techniques in Asia.
