3D automotive aerodynamics

Goal

Compare geometry-aware surrogates on realistic vehicle shapes and production-scale meshes.

Candidate datasets

DrivAerNet++, DrivAerML, and ShapeNet-Car.

Recommended models

GINO, DoMINO, Transolver, UPT, P3D, AeroTransformer, and graph baselines.

Required reporting

Surface-only versus full-volume prediction, drag and pressure coefficients, mesh statistics, wall treatment, memory, throughput, and geometry-OOD performance.

References

DrivAerNet++; DrivAerML; GINO; DoMINO; AeroTransformer.