Luminary Unveils SHIFT-Crash: First Physics-Based AI Model for Full-Vehicle Crash Prediction

At the 2026 SAE World Congress in Detroit, U.S., physical AI company Luminary unveiled an AI model named SHIFT-Crash, capable of predicting full-vehicle crash responses—including deformation and stress field distributions—in just seconds, dramatically faster than the several hours required by traditional methods. Trained on 5,000 crash simulations of the 2010 Toyota Yaris, SHIFT-Crash is the first learning-based surrogate model that captures crash dynamics across an entire vehicle and a multidimensional parametric design space. It can directly generate spatially resolved stress field predictions from vehicle design parameters. Unlike conventional finite element method (FEM) approaches that rely on high-performance computing clusters, SHIFT-Crash functions as a reusable AI model, retaining and reapplying crash physics knowledge across different vehicle programs—eliminating the need to rerun simulations from scratch for each new design. Traditional FEM simulations typically take 10 to 12 hours for a single NHTSA NCAP frontal crash test at 56 km/h. Moreover, due to their inability to transfer learned knowledge, detailed crash safety analyses are often delayed until late in the development cycle, severely limiting opportunities for design optimization.