DrivAerNet++: Multimodal CFD Automotive Dataset

• DrivAerNet++ is a large-scale, multimodal automotive CFD dataset featuring 8,000 simulated car designs with detailed 3D meshes, aerodynamic coefficients, and full-field flow data.
• It supports diverse engineering tasks such as drag-coefficient regression, surface-pressure prediction, volumetric flow inference, and controllable geometry generation with validated performance metrics.
• The dataset’s standardized benchmark tasks and rich parametric design space make it a valuable resource for data-driven automotive aerodynamics research and inverse design optimization.

DrivAerNet++ is a large-scale, multimodal, high-fidelity dataset for automotive aerodynamics built around industry-standard parametric car geometries and computational fluid dynamics (CFD) simulations. It was introduced as an extension of the earlier DrivAerNet resource, expanding both geometric diversity and physical supervision through 3D meshes, parametric models, aerodynamic coefficients, surface and volumetric flow fields, point clouds, and part annotations (Elrefaie et al., 2024). Across subsequent work, it has functioned both as a dataset in its own right and as a benchmark substrate for drag-coefficient regression, surface-pressure prediction, wall-shear-stress estimation, volumetric flow inference, controllable geometry generation, operator learning, and active 3D reconstruction (Chen et al., 19 Mar 2025).

1. Origin, scope, and defining characteristics

DrivAerNet++ was presented as “the largest and most comprehensive multimodal dataset for aerodynamic car design,” comprising 8,000 diverse car designs modeled with high-fidelity CFD simulations and more than 39 TB of publicly available engineering data (Elrefaie et al., 2024). The dataset substantially extends the original DrivAerNet, which contained 4,000 detailed 3D car meshes focused on the DrivAer fastback family, by broadening the configuration space to multiple body styles, wheel types, and underbody variants (Elrefaie et al., 2024).

Its core novelty is multimodality. Each entry may include detailed 3D meshes, parametric models, segmented parts, aerodynamic coefficients, full-domain 3D flow fields, surface fields, and point clouds (Elrefaie et al., 2024). The dataset is therefore not restricted to scalar aerodynamic labels such as drag; it also supports dense field-learning tasks and geometry-centric learning tasks. In the benchmarking literature built on top of it, DrivAerNet++ is repeatedly described as a high-fidelity automotive CFD corpus that supports prediction directly from 3D geometry, bypassing 2D rendering and, in some methods, signed distance function pre-processing (He et al., 11 Apr 2025).

The dataset was designed around practical aerodynamic variation rather than abstract shape diversity. It includes fastback, notchback, and estateback variants, multiple wheel options, smooth and detailed underbodies, mirrors, and related external details (Elrefaie et al., 2024). Later studies emphasize that it is unusual among public automotive datasets in explicitly including tires, chassis, wheels, and underbody structure, which materially affect near-ground and wheel-wake aerodynamics (He et al., 11 Apr 2025). This emphasis on configuration realism, rather than only gross body shape, is central to why DrivAerNet++ became a reference benchmark for learning-based automotive aerodynamics.

A recurrent point in later papers is that DrivAerNet++ is both a dataset and a family of benchmark settings. Some works refer to “over 8,000” designs, some to 8,121 designs, and some to 8,150 steady-state CFD simulations (Chen et al., 19 Mar 2025, Zheng et al., 24 Feb 2026, Elrefaie et al., 25 Nov 2025). This suggests evolving curation or benchmark-specific preprocessing, but the canonical introduction defines the resource at 8,000 designs and more than 39 TB of data (Elrefaie et al., 2024).

2. Geometric design space and data modalities

The geometric basis of DrivAerNet++ is the DrivAer parametric car model family, expanded into a broader configuration space through ANSA-based morphing and parameter sampling (Elrefaie et al., 2024). The dataset contains 26 geometric design parameters in the published DrivAerNet++ description, with examples including car width, length, roof height, greenhouse angle, diffusor angle, ramp angle, trunklid angle, rear-window inclination and length, windscreen inclination and length, pillar thicknesses, fender arch offset, door-handle position and thickness, trunklid curvature, front-bumper curvature, underbody type, rear-body category, and wheel type (Elrefaie et al., 2024). In the generative-modeling literature, these same 26 parameters are described as “human-interpretable automotive design parameters,” including quantities such as length, width, roof height, ramp angle, front bumper curvature, diffuser angle, and car length (Nehme et al., 26 Oct 2025).

The main geometric assets are STL meshes of car bodies and wheels, described as simulation-ready and watertight (Elrefaie et al., 2024). Surface discretization is industrial in scale: the original DrivAerNet++ paper reports CFD meshes with car-surface cells on the order of $\approx 500\text{–}750\mathrm{k}$, while later benchmark papers refer to more than 500,000 surface points per vehicle surface mesh (Elrefaie et al., 2024, Zheng et al., 24 Feb 2026). Point-cloud variants are also distributed at multiple resolutions, including 5k, 10k, 100k, 250k, and 500k surface samples (Elrefaie et al., 2024).

The dataset’s physical modalities are unusually rich for an open engineering dataset. Per design, it may provide aerodynamic coefficients such as $C_D$, $C_L$, front and rear lift coefficients, and moment coefficients; full 3D velocity and pressure fields; turbulence quantities $k$ and $\omega$; surface pressure; wall shear stress $\tau_w$; and pressure coefficient $C_p$ maps (Elrefaie et al., 2024). Some downstream papers use only scalar $C_d$ labels for surrogate regression (Singh et al., 5 Jan 2026), while others exploit dense surface pressure, wall shear stress, or full 3D velocity supervision (Chen et al., 19 Mar 2025, Zheng et al., 24 Feb 2026).

Annotation breadth also matters. The original dataset release reports 29 distinct part labels for segmentation and classification (Elrefaie et al., 2024), whereas later field-prediction work describes 20-part annotations used for per-part drag analyses (Zheng et al., 24 Feb 2026). This again indicates benchmark-layer adaptation rather than a single immutable export. A plausible implication is that different derivative releases or task-specific preprocessed versions expose different subsets of the original annotation structure.

3. CFD generation, aerodynamic quantities, and validation

DrivAerNet++ is grounded in steady incompressible Reynolds-averaged Navier–Stokes simulation with the $k\text{–}\omega$ SST turbulence model in OpenFOAM (Elrefaie et al., 2024). The canonical dataset paper specifies OpenFOAM v11, simpleFoam / incompressibleFluid, potentialFoam initialization, snappyHexMesh meshing, prism layers around the car, wheels, and ground, and detailed near-body refinement (Elrefaie et al., 2024). Operating conditions are standardized: inlet velocity $U_{\infty} = 30 \,\mathrm{m/s}$, air density $C_D$0, kinematic viscosity $C_D$1, and Reynolds numbers spanning $C_D$2 to $C_D$3 depending on vehicle length (Elrefaie et al., 2024).

Boundary conditions are configured for external automotive aerodynamics. The inlet uses uniform velocity, the outlet uses a pressure outlet, top and lateral boundaries use slip, the ground is a moving wall, and the wheels use rotating-wall kinematics,

$C_D$4

Symmetry is used for symmetric designs (Elrefaie et al., 2024). These details are important because later papers often cite DrivAerNet++ as realistic precisely because it models rotating wheels and moving ground rather than static stand-alone bodies (Chen et al., 19 Mar 2025, Elrefaie et al., 25 Nov 2025).

The aerodynamic quantities used throughout the ecosystem follow standard definitions. The dataset paper defines drag, lift, pressure coefficient, and Reynolds number as

$C_D$5

$C_D$6

These definitions are inherited or reused by later surrogate-model papers, including drag-prediction and field-prediction benchmarks (Elrefaie et al., 2024, He et al., 11 Apr 2025).

Validation against experimental references is a defining feature of the dataset release. For baseline DrivAer geometries, the reported drag deviations versus experiment are 2.88% for fastback smooth, 2.18% for fastback detailed, 4.88% for notchback, and 4.11% for estateback (Elrefaie et al., 2024). The authors therefore position the dataset as high-fidelity for academic surrogate modeling while also acknowledging that it remains steady RANS rather than higher-cost transient hybrid RANS–LES or LES. This is a central interpretive point: DrivAerNet++ is not raw reality, but a standardized, validated CFD approximation of reality at scale.

4. Benchmark tasks and learning paradigms

DrivAerNet++ has become a benchmark hub rather than a single-task dataset. The earliest benchmarking emphasis was scalar drag-coefficient prediction from 3D geometry. The dataset paper reports that PointNet, GCNN, and RegDGCNN achieve $C_D$7 values of 0.643, 0.596, and 0.641 respectively on the mixed-category DrivAerNet++ test set of 1,200 designs, showing that generalization across diverse configurations is substantially harder than within a single fastback family (Elrefaie et al., 2024). TripNet later raised drag-prediction performance to $C_D$8 with $C_D$9 and $C_L$0 on a 1,200-design unseen test set (Chen et al., 19 Mar 2025). A slice-based surrogate model based on PointNet2D and Bi-LSTM reported $C_L$1, $C_L$2, and inference time of approximately 0.025 seconds per sample on a GeForce RTX 4060 Laptop GPU (Singh et al., 5 Jan 2026). Transformer-based drag prediction with DrivAer Transformer yielded $C_L$3 and $C_L$4 on a 70/15/15 split, outperforming PointNet, GCNN, and RegDGCNN under that evaluation protocol (He et al., 11 Apr 2025).

A second benchmark axis is dense surface-field prediction. For surface pressure on DrivAerNet++, TripNet reports $C_L$5 and $C_L$6, while AdaField improves these to $C_L$7 and $C_L$8 respectively (Chen et al., 19 Mar 2025, Zou et al., 12 Jan 2026). GA-Field further reports normalized-pressure performance of $C_L$9, $k$0, $k$1, and $k$2, alongside wall-shear-stress magnitude prediction with $k$3 and $k$4 (Zheng et al., 24 Feb 2026). GIST later reports state-of-the-art surface-pressure prediction on DrivAerNet++ with 3.63 MSE and 18.60% relative $k$5 error, improving over TripNet, FigConvNet, Transolver, and RegDGCNN (Rigotti et al., 17 Mar 2026).

A third axis is volumetric-flow prediction. TripNet reports 3D velocity-field results on DrivAerNet++ with relative $k$6 errors of 7.15% for $k$7, 28.97% for $k$8, 31.12% for $k$9, and 6.88% for velocity magnitude $\omega$0 (Chen et al., 19 Mar 2025). GA-Field improves these figures, reporting for instance $\omega$1 relative $\omega$2 improvement from 10.71 to 9.91 and velocity-magnitude relative $\omega$3 improvement from 10.39 to 9.65 versus TripNet under its evaluation setting (Zheng et al., 24 Feb 2026).

DrivAerNet++ has also become a benchmark for geometry generation and controllable design. In LAMP, the dataset is treated as a large-scale parametric car corpus with approximately 8,000 distinct geometries, high-resolution meshes, CFD-based aerodynamic coefficients, and 26 interpretable design parameters (Nehme et al., 26 Oct 2025). In that setting, the benchmark covers controlled interpolation, within-range extrapolation, large-range extrapolation beyond the dataset bounds, and performance-guided optimization subject to geometric constraints. LAMP reports safe extrapolation by up to 100% parameter difference beyond training ranges and $\omega$4-targeted optimization under fixed parameters on 100 random test examples (Nehme et al., 26 Oct 2025). This establishes DrivAerNet++ not merely as a supervised regression dataset but as a design-space substrate for inverse problems.

5. Standardized evaluations, cross-category generalization, and benchmark infrastructure

As the ecosystem matured, DrivAerNet++ ceased to be only a dataset release and became a standardized benchmark platform. CarBench formalizes this role by evaluating eleven architectures on DrivAerNet++, focusing on surface-pressure prediction from geometry alone at a fixed freestream condition (Elrefaie et al., 25 Nov 2025). In that benchmark, the official split is given as 5,819 train, 1,177 validation, and 1,154 test samples; evaluation is performed on 10,000 sampled surface points per car, with metrics reported in denormalized physical units of kinematic pressure (Elrefaie et al., 25 Nov 2025). The benchmark also introduces stratified paired bootstrap with $\omega$5 resamples to quantify uncertainty in reported scores.

CarBench highlights a ranking among strong recent models. On the 1,154-sample test set, AB-UPT achieves $\omega$6 and $\omega$7, followed by TransolverLarge, Transolver, Transolver++, and TripNet (Elrefaie et al., 25 Nov 2025). The benchmark also reports memory and latency trade-offs, showing, for example, that AB-UPT uses 0.27 GB and 30.6 ms per sample, while NeuralOperator reaches 466 samples/s at the cost of lower predictive accuracy (Elrefaie et al., 25 Nov 2025). This benchmarking layer is important because raw numbers reported in isolated papers are often not directly comparable due to differing splits, normalization schemes, and output representations.

Cross-category generalization is one of the defining difficulties of DrivAerNet++. GA-Field evaluates three out-of-distribution protocols in which models are trained on two body archetypes and tested on the held-out third archetype (Zheng et al., 24 Feb 2026). In the most difficult setting, training on Estateback and Notchback and testing on Fastback, GA-Field reports $\omega$8 and $\omega$9, improving over Transolver and Transolver++ but still leaving a substantial generalization gap (Zheng et al., 24 Feb 2026). CarBench reaches a similar conclusion from zero-shot transformer baselines: the hardest direction is again $\tau_w$0Estateback + Notchback$\tau_w$1 Fastback, with relative $\tau_w$2 around 0.49–0.50 and $\tau_w$3 (Elrefaie et al., 25 Nov 2025). The consistent message is that body-style diversity remains a genuine distribution-shift challenge even at current dataset scale.

A common misconception is that the sheer size of DrivAerNet++ removes the need for task-specific evaluation protocols. The benchmark literature shows the opposite. Results depend strongly on whether one predicts scalar $\tau_w$4, pressure coefficient maps, normalized pressure, kinematic pressure, wall shear stress, or 3D velocity; whether evaluation is performed on sparse sampled points or full meshes; and whether the task is in-distribution interpolation or cross-category generalization (Chen et al., 19 Mar 2025, Elrefaie et al., 25 Nov 2025). DrivAerNet++ therefore functions less as a single benchmark than as an extensible benchmark suite.

6. Scientific uses, limitations, and open issues

The dataset supports a broad spectrum of scientific and engineering workflows. The original release emphasizes data-driven design optimization, generative modeling, surrogate modeling of aerodynamic performance, CFD acceleration, geometric classification, segmentation, and 3D surface-field prediction (Elrefaie et al., 2024). LAMP uses it for controllable and safe 3D generation under explicit geometric constraints (Nehme et al., 26 Oct 2025). OTNO uses the DrivAerNet family for instance-dependent optimal-transport geometry embedding and latent-manifold operator learning for drag prediction (Li et al., 26 Jul 2025). FluidGaussian uses DrivAerNet++ meshes as ground-truth vehicle geometries in physics-aware active 3D reconstruction, where aerodynamic surface quality is probed by simulation-induced uncertainty during view selection (Liu et al., 22 Mar 2026).

At the same time, the literature repeatedly identifies structural limitations. First, style diversity is not unrestricted. DrivAer Transformer notes that many shapes are obtained by morphing four base templates, which can limit style diversity and generalization to OEM-specific designs (He et al., 11 Apr 2025). Second, the CFD supervision is standardized but narrow: all mainstream benchmark papers use steady-state RANS settings around $\tau_w$5, and crosswind, transient effects, and broader operating envelopes are generally absent from the public benchmark regime (Elrefaie et al., 2024, Zheng et al., 24 Feb 2026). Third, while the dataset is high fidelity, some downstream papers stress that robust prediction of subtle geometry changes, add-on sensors, spoilers, and accessories remains difficult under domain shift (He et al., 11 Apr 2025).

There are also practical ambiguities. Different papers cite different sample counts and official splits, and not all benchmark layers expose the same modalities or preprocessing conventions (Zheng et al., 24 Feb 2026, Elrefaie et al., 25 Nov 2025). This suggests that replication requires attention to the exact benchmark release rather than only the dataset name. Another recurring issue is output normalization. Several surface-pressure methods normalize pressure by subtracting the dataset mean $\tau_w$6 and dividing by the standard deviation $\tau_w$7 (Zheng et al., 24 Feb 2026, Zou et al., 12 Jan 2026), while CarBench evaluates in denormalized physical units of kinematic pressure (Elrefaie et al., 25 Nov 2025). These are not interchangeable reporting conventions.

Future directions in the literature converge on several themes. Proposed extensions include broader vehicle classes such as SUVs and additional OEM styles, multi-objective labels such as lift and pressure distributions, richer operating conditions, uncertainty quantification, domain adaptation for accessories and add-on components, physics-informed constraints, and multi-fidelity learning beyond steady RANS (He et al., 11 Apr 2025, Zou et al., 12 Jan 2026, Elrefaie et al., 2024). This suggests that DrivAerNet++ is best understood not as a finished benchmark, but as the current reference point in a rapidly expanding line of data-driven automotive aerodynamics research.