Update (26.05.2025): We've updated our benchmarking with extensive results comparing TripNet with Nvidia's FigConvNet, Transolver, and RegDGCNN (paper). We're also releasing new data modalities for DrivAerNet++ soon, including photorealistic renderings, sketches, 2D slices, SDFs, and more.
Update (18.04.2025): We are open-sourcing the RegDGCNN pipeline for surface field prediction on 3D car meshes from DrivAerNet++.
Update (01.04.2025): We are excited to share our new paper on AI Agents in Engineering Design, where we introduce a multi-agent framework that leverages vision-language models (VLMs), large language models (LLMs), and geometric deep learning to accelerate the car design process—from concept sketching to CAD modeling, meshing, and aerodynamic simulation. This system enables real-time interaction between designers, engineers, generative AI models, and tools like Blender, OpenFOAM, and ParaView.
Update (19.11.2024): DrivAerNet++ has been accepted to NeurIPS 2024! The full dataset is now released on Harvard Dataverse. Please note the (CC BY-NC 4.0) license terms, as outlined in the License section.
Update (11.09.2024): Due to the overwhelming interest and numerous inquiries from industry partners, we are excited to announce that we are now offering commercial licensing options for the DrivAerNet and DrivAerNet++ datasets. Please refer to the DrivAerNet/DrivAerNet++ Commercial Inquiry section.
📄 DrivAerNet++ Paper (NeurIPS'24) | 📄 DrivAerNet Paper (JMD) | 🎥 Video Summary
We present DrivAerNet++, the largest and most comprehensive multimodal dataset for aerodynamic car design. DrivAerNet++ comprises 8,000 diverse car designs modeled with high-fidelity computational fluid dynamics (CFD) simulations. The dataset includes diverse car configurations such as fastback, notchback, and estateback, with different underbody and wheel designs.
Design parameters for the generation of the DrivAerNet++ dataset. Several geometric parameters with significant impact on aerodynamics were selected and varied within a specific range. These parameter ranges were chosen to avoid values that are either difficult to manufacture or not aesthetically pleasing.
DrivAerNet++ covers all conventional car designs. The dataset encompasses various underbody and wheel designs to represent both internal combustion engine (ICE) and electric vehicles (EV). This extensive coverage allows for comprehensive studies on the impact of geometric variations on aerodynamic performance. By including a diverse set of car shapes, DrivAerNet++ facilitates the exploration of aerodynamic effects across different vehicle types, supporting both academic research and industrial applications.
 |
 |
Each 3D car geometry is parametrized with 26 parameters that completely describe the design. To create diverse car designs, we used two main morphing methods: morphing boxes and direct morphing. For a detailed description of the design parameters, their ranges, lower and upper bounds, please refer to the paper.
Dataset diversity and shape variation are crucial for developing robust deep learning models in aerodynamic car design. By providing a wide range of car shapes and configurations with high-fidelity CFD, DrivAerNet++ enables models to generalize better, supports exploration of unconventional designs, and enhances understanding of how geometric features impact aerodynamic performance.
- Parametric Models
Parametric car models with structured tabular design parameters, enabling controlled design variation and sensitivity studies. - Volumetric Fields
Full 3D CFD simulation data (e.g., pressure, velocity, turbulence) in the flow domain around each vehicle. - Surface Fields
Surface-level quantities such as pressure coefficient (Cp) and wall shear stress (WSS), mapped directly onto the car body. - Streamlines
Flow visualization data illustrating streamlines around the car geometry, capturing wake structure and aerodynamic behavior. - Point Clouds
Dense and sparse point cloud representations derived from surface meshes. - Meshes
High-resolution 3D surface triangulations for geometry-based neural networks and meshing pipelines. - Aerodynamic Coefficients
Global performance metrics such as drag coefficient (Cd), lift coefficient (Cl), and moment coefficients, computed via CFD. - Annotations
Per-part semantic labels for each car, enabling part-aware learning and geometric reasoning.
- Renderings
High-quality photorealistic 2D renderings from multiple views, useful for multimodal learning and image-based supervision. - Sketches
Hand-drawn style sketches for vision-based tasks and generative models. - 2D Slices
Planar field extractions (2d silhouettes, 2d mesh, pressure, and velocity). - Signed Distance Fields (SDF)
SDF representations of car shapes for occupancy modeling and implicit surface learning. - Deformations
Simulation outputs under crash or pressure conditions for learning physical response under impact or force.
In addition to the CFD simulation data, our dataset includes detailed annotations for various car components (29 labels), such as wheels, side mirrors, and doors. These annotations are instrumental for a range of machine learning tasks, including classification, semantic segmentation, and object detection. The comprehensive labeling can also facilitate automated CFD meshing processes by providing precise information about different car components. By incorporating these labels, our dataset enhances the utility for developing and testing advanced algorithms in automotive design and analysis.
Car design is not just an engineering challenge — it's an art form.
To bridge the gap between conceptual creativity and computational design, we extend DrivAerNet++ with new modalities: 2D hand-drawn sketches and photorealistic renderings.
🖼️ Engineers and designers now have access to both engineering data (meshes, CFD simulations, surface fields) and design inputs (sketches and styled renderings) — in one unified dataset.
 |
 |
🔍 For details, check out our recent Design Agents paper: AI Agents in Engineering Design
Running the high-fidelity CFD simulations for DrivAerNet++ required substantial computational resources. The simulations were conducted on the MIT Supercloud, leveraging parallelization across 60 nodes, totaling 2880 CPU cores, with each CFD case using 256 cores and 1000 GBs of memory. The full dataset requires 39 TB of storage space. The simulations took approximately 3 × 10⁶ CPU-hours to complete.
DrivAerNet++ supports a wide array of machine learning applications, including but not limited to:
- 🚀 Data-driven design optimization: Optimize car designs based on aerodynamic performance.
- 🧠 Generative AI: Train generative models to create new car designs based on performance or aesthetics.
- 🎯 Surrogate models: Predict aerodynamic performance without full CFD simulations.
- 🔥 CFD simulation acceleration: Speed up simulations using machine learning and multi-GPU techniques.
- 📉 Reduced Order Modeling: Create data-driven reduced-order models for efficient & fast aerodynamic simulations.
- 💾 Large-Scale Data Handling: Efficiently store and manage large datasets from high-fidelity simulations.
- 🗜️ Data Compression: Implement high-performance lossless compression techniques.
- 🌐 Part and shape classification: Classify car categories or components to enhance design analysis.
- 🔧 Automated CFD meshing: Automate the meshing process based on car components to streamline simulations.
The DrivAerNet++ dataset is hosted under the CC BY-NC 4.0 license on Harvard Dataverse. The dataset is structured into four subsets:
- 3D Meshes: Parametric car geometries in STL format.
- Pressure: Surface pressure field data.
- Wall Shear Stress: Aerodynamic wall shear stress distributions.
- CFD (Full CFD Domain): Complete volumetric CFD simulation data.
We provide instructions on how to use Globus and to download the dataset. Please feel free to reach out if you encounter any problems/issues.
DrivAerNet++ serves as a valuable benchmark dataset due to its size and diversity. It provides extensive coverage of various car designs and configurations, making it ideal for testing and validating machine learning models in aerodynamic design. We provide the train, test, and validation splits in the following folder: train_val_test_splits.
Drag values for the 8k car designs can be found Here, and the frontal projected areas Here.
Researchers and industry practitioners can submit their models to the leaderboard to compare performance against state-of-the-art baselines. The benchmark promotes transparency, reproducibility, and innovation in AI-driven aerodynamic modeling.
For submission guidelines and current rankings, visit DrivAerNet++ Leaderboard.
DrivAerNet++ stands out as the largest and most comprehensive dataset in the field of car design.
DrivAerNet has been integrated into leading Scientific Machine Learning (SciML) frameworks, enabling accelerated aerodynamic predictions, surrogate modeling, and generative AI-based design optimization.
DrivAerNe is integrated into NVIDIA Modulus, supporting advanced deep learning models for CFD acceleration:
DrivAerNet++ was featured in the IJCAI 2024 competition - Rapid aerodynamic drag prediction for arbitrary vehicles using deep learning. The dataset is also integrated into PaddleScience, Baidu’s SciML toolkit for physics-based AI.
🔗 IJCAI 2024 Competition 🔗 PaddleScience DrivAerNet Example 🔗 PaddleScience DrivAerNet++ Example
We welcome contributions to improve the dataset or project. Please submit pull requests for review.
Maintained by the DeCoDE Lab at MIT. Report issues via GitHub issues.
- Tutorials: Link
To replicate the code and experiments from the first version of DrivAerNet, please refer to the folder: DrivAerNet_v1.
Strict Licensing Notice: DrivAerNet/DrivAerNet++ is released under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0) and is exclusively for non-commercial research and educational purposes. Any commercial use—including, but not limited to, training machine learning models, developing generative AI tools, creating software products, running new simulations using the provided geometries or any derived geometries, or other commercial R&D applications—is strictly prohibited. Unauthorized commercial use of DrivAerNet/DrivAerNet++, or any derived data, will result in enforcement by the MIT Technology Licensing Office (MIT TLO) and may carry legal consequences. The code is distributed under the MIT License.
If you are interested in the commercial use of the DrivAerNet or DrivAerNet++ datasets, please contact Mohamed Elrefaie ([email protected]) and Faez Ahmed ([email protected]) with the subject line: "DrivAerNet Commercial Inquiry".
To cite this work, please use the following reference:
@inproceedings{NEURIPS2024_013cf29a,
author = {Elrefaie, Mohamed and Morar, Florin and Dai, Angela and Ahmed, Faez},
booktitle = {Advances in Neural Information Processing Systems},
editor = {A. Globerson and L. Mackey and D. Belgrave and A. Fan and U. Paquet and J. Tomczak and C. Zhang},
pages = {499--536},
publisher = {Curran Associates, Inc.},
title = {DrivAerNet++: A Large-Scale Multimodal Car Dataset with Computational Fluid Dynamics Simulations and Deep Learning Benchmarks},
url = {https://proceedings.neurips.cc/paper_files/paper/2024/file/013cf29a9e68e4411d0593040a8a1eb3-Paper-Datasets_and_Benchmarks_Track.pdf},
volume = {37},
year = {2024}
}To cite the first version of DrivAerNet, please use the following reference:
@proceedings{10.1115/DETC2024-143593,
author = {Elrefaie, Mohamed and Dai, Angela and Ahmed, Faez},
title = {DrivAerNet: A Parametric Car Dataset for Data-Driven Aerodynamic Design and Graph-Based Drag Prediction},
volume = {Volume 3A: 50th Design Automation Conference (DAC)},
series = {International Design Engineering Technical Conferences and Computers and Information in Engineering Conference},
pages = {V03AT03A019},
year = {2024},
month = {08},
doi = {10.1115/DETC2024-143593},
url = {https://doi.org/10.1115/DETC2024-143593},
eprint = {https://asmedigitalcollection.asme.org/IDETC-CIE/proceedings-pdf/IDETC-CIE2024/88360/V03AT03A019/7402927/v03at03a019-detc2024-143593.pdf},
}