Introducing NVIDIA Halos SAFE: A Safety Assurance Framework for Autonomous Driving

NVIDIA Halos SAFE, the Safety Assurance Framework for Evaluation, connects road testing, simulation, and fleet data into a single safety loop. The announcement is framed around the Mercedes-Benz CLA, powered by NVIDIA Hyperion and Halos, which was named Euro NCAP's Best Performer of 2025 with the year's highest overall safety score. The session walks through how continuous safety validation is carried out from real-world test tracks to simulation and back into development.

■ Overview
- The role of evidence in proving vehicle safety, from L2 driver assistance to L4 robotaxis
- NCAP test suites and safety evaluations across worldwide markets
■ The Safety Loop
- Capturing driving scenarios on the test track to establish ground truth
- Reconstructing scenes with NVIDIA Omniverse NuRec and reproducing them in simulation
- Comparing simulated vehicle responses against track measurements
- Reusing validated simulations to test new software releases
■ Hardware and Real-World Testing
- Hardware-in-the-loop simulation with NVIDIA DRIVE AGX Thor
- Real cars tested against virtual hazards using simulated sensor inputs in real time
- Safety-target measurement, feeding results back into development
■ Scenario Expansion
- NVIDIA Cosmos generating thousands of offline variations in lighting, weather, road surfaces, and traffic
- Continuous validation as evidence for certification across markets

This is useful for engineers, safety and certification professionals, and anyone evaluating autonomous driving technology. Viewers can expect to understand the structure of a modern safety validation workflow and how each tool fits into it, then apply that framework when assessing their own testing or compliance plans.