NVIDIA Halos SAFE Autonomous Vehicle Safety Evaluation From Road Testing to Simulation

The Mercedes-Benz CLA, powered by NVIDIA Hyperion and Halos, was Euro NCAP's best performer of 2025, with the highest overall safety score of the year. Safety isn't easy, and it needs to be proven. That takes evidence. Today, we're announcing NVIDIA Halo Safe, the safety assurance framework for evaluation. It connects road testing, simulation, and fleet data in one safety loop. Halo Safe provides test suites for ANCAP testing worldwide and safety evaluations from L2 driver assistance to L4 robo-taxis. It starts in the real world. We capture a sample of driving scenarios on the test track and measure exactly what happened. That becomes the ground truth. NVIDIA Omniverse Neuregg reconstructs the scene. We run the same test and simulation and compare the vehicle's response with track measurements. Once simulation results match real-world scenarios, they can be reused to test new software releases. We test the complete system, including NVIDIA DRIVE AGX-THOR and a hardware-in-the-loop simulator to check that improvements seen in simulation carry through to the hardware. We test the real car against virtual hazards. The test car receives simulated sensor inputs in real time, creating obstacles and edge cases that aren't physically there. The vehicle sees them, reacts to them, and lets us test dangerous situations safely. Every result is measured against a safety target so we know what passed, what failed, and why. What we learn goes back into development, And the next version of the software is tested again to test how the car responds as real-world conditions vary. NVIDIA Cosmos generates thousands of variations offline in lighting, weather, road surfaces, and traffic. This is NVIDIA Halo Safe. Continuous safety validation. Evidence for certification across markets so every new version can be proven safe and deployed with confidence around the world.