NVIDIA Research’s CANTO aims to predict aerodynamic fields directly from CAD

A NVIDIA Research researcher introduced CANTO, a system designed to predict aerodynamic fields from native CAD and use gradients to improve designs. They report 20% lower surface-pressure relative L₂ error than AB-UPT on HiLiftAeroML. On AhmedML, they report 4–20% lower drag than the best eligible dataset designs under matching volume and lift constraints, with the result validated using CFD.