MIT has created an open-source database of 8,000 cars to give designers a head start

Engineers at the Massachusetts Institute of Technology decided to prove that artificial intelligence can create significant progress in the design of cars, especially in the realm of aerodynamics. As in other industries, AI can comb through vast amounts of data, separate out what’s useful and what’s not, and design a car accordingly. It’s not great news if you’re currently studying, say, automotive design, but MIT claims that artificial intelligence can make cars more efficient.

The AI ​​tools needed to design a car from scratch with aerodynamics in mind already exist, but until recently the data was either private, decentralized, or non-existent. That’s where MIT engineers stepped in. They created a dataset called DrivAerNet++ which includes more than 8000 different car designs presented in 3D form. Each design includes aerodynamic information about the car it represents, based on simulations.

Achieving this was easier said than done. The engineers started with a few basic 3D models provided by Audi and BMW in 2014, which they divided into three categories, respectively: fastback, cam and bush. (MIT obviously leaves the convertible design up to the humans.) Once those files were uploaded and ready to go, the MIT team made small changes to 26 different parameters (including length, various parts of the underbody, and the angle of the windshield) and saved each modification as a separate car . That’s how they got over 8,000 designs.

We didn’t go through the 8,000 designs, but we did spot some familiar-looking cars in a few released by MIT. One is clearly based on the E91-gen BMW 3 Series Touring, albeit with trippy-like acid colors and a few visual tweaks. The other looks like an Audi A4 Avant, so at least our future AI overlords will let us keep driving wagons (probably in exchange for a pencil and piece of paper as a sacrifice).

MIT hopes that engineers around the world will use this giant amount of data to train artificial intelligence models to design a car. In turn, anyone who has access to these tools, including youpotentially—could develop a station wagon faster than, say, General Motors, and theoretically at lower cost.

“Often in developing a car, the upfront process is so expensive that manufacturers can only slightly tweak the car from one version to the next,” said Faiz Ahmed, associate professor of mechanical engineering at MIT. “But if you have large data sets where you know the performance of each design, now you can train a machine learning model to iterate quickly so you have a better chance of getting a better design.”

Can industry teach AI sculpting a car that is both aerodynamic and visually appealing? Will artificial intelligence eventually be able to take into account other factors that influence the car design process, such as manufacturing feasibility and regulatory crumple zones? And heck, won’t the “pre-AI model” become a selling point in the enthusiast community one day? Time will tell – and we may not have to wait long to find out.

Any tips? Send them tips@thedrive.com