Volvo trains its AI autopilot by showing 3D scenes based on real-life images

Volvo Cars has announced the use of artificial intelligence and virtual simulations to improve the safety of its cars. ADAS (driver assistance systems) can now learn not only from real-life accidents, but also from thousands of simulated scenarios that Volvo creates using advanced 3D technology.

Volvo analyzes data on emergency braking, sharp maneuvers and driver intervention collected by the car’s sensors. To do this, it uses a hybrid method of representing 3D data, also known as Gaussian splatting. This technology allows you to create realistic 3D scenes based on real images. In these virtual environments, you can change the behavior of other road users, testing the system under different conditions.

According to the developers, this significantly speeds up the training of the autopilot: what previously took months to collect data in rare and complex situations, now they can be tested in a matter of days.

Volvo is combining real-world testing and virtual simulation to create the safest cars. The project is being carried out in collaboration with Zenseact (Volvo’s AI subsidiary) and with support from the Wallenberg AI, Autonomous Systems and Software Program (WASP). The research is also part of a PhD program at leading universities in Sweden.

In the future, such technologies could make safety systems even more adaptive and effective – especially in unpredictable situations.