UniDrag: Unified Multi-Field Prediction and Robust Shape Optimization for Vehicle Aerodynamics

High-fidelity vehicle aerodynamics analysis is bottlenecked by costly CFD simulations. Neural surrogates accelerate prediction but lack inverse design capabilities, while existing generative optimization methods suffer from unstable convergence and frequent engineering constraint violations. We present UniDrag, a unified framework bridging multi-field aerodynamic prediction with robust differentiable shape optimization. Given a vehicle geometry, UniDrag predicts surface pressure, volume flow fields, drag coefficient $C_d$, and a streamwise build-up profile localizing drag contributions. Our architecture combines enhanced Physics-Sliced Attention (ePSA), Gated Expert Routing, and Modality-Protected Learning to prevent negative transfer across output modalities. At deployment, the frozen surrogate enables gradient-based optimization via Free-Form Deformation with engineering constraints. We introduce Expectation-over-Transformation to prevent adversarial exploitation of surrogate fragility. We curate a large-scale dataset of 15,000 vehicle geometries spanning four body types with GPU-accelerated LBM simulations. On this benchmark, UniDrag achieves $C_d$ prediction $R^2$ of 0.937 (+7.6\% over baselines) and 13.7\% mean CFD-verified drag reduction with 100\% success rate and only 21.3~mm average displacement (0.46\% vehicle length).

Making a car "slip" through the air more smoothly is one of the most important ways to cut fuel consumption and to extend the driving range of electric vehicles. To evaluate a candidate car shape, engineers traditionally rely on detailed airflow simulations that take hours each, so only a handful of designs can be explored before a model enters production. Recent AI systems can predict the airflow around a car much faster than these simulations, but they can only score a given shape — they cannot tell a designer how to improve it. Other AI systems can generate brand-new shapes from scratch, yet the cars they produce are often left-right asymmetric, oddly shaped, or simply impossible to manufacture. This paper introduces UniDrag, a single AI system that does both jobs at once. Given a car, it predicts the surrounding airflow and pinpoints where on the body the air resistance is coming from, and then it proposes small, smooth, factory-feasible changes to the shape that reduce that resistance. Crucially, UniDrag respects the rules real engineers care about: the body must stay symmetric, wheels and windows must remain untouched, and the surface must stay smooth enough for a stamping die. To train and test the system, we built a dataset of 15,000 detailed car models spanning sedans, SUVs, MPVs, and sports cars. When the refined shapes are re-checked with full physics simulations, UniDrag reliably reduces aerodynamic drag by an average of 13.7%, on every test vehicle, while moving the body surface by only a few millimetres. The result is a practical tool that helps designers find more energy-efficient car shapes early in the design process, long before any prototype is built.