With a feel for physics, AI models simulate a wider range of real-world scenarios

Simulating physics in AI has been slow and data-heavy, limiting how engineers test vehicles, robots, and designs. MIT and Tsinghua University researchers developed GeoPT, a pre-training approach that teaches AI physics by simulating tiny particles interacting with 3D shapes. GeoPT reaches peak performance twice as fast, needs 60% less labeled data, and accurately simulates wind, crashes, waves, and light across complex industrial scenarios.