NVIDIA Halos SAFE Connects Road Tests, Simulation, and Hardware for AV Evaluation
NVIDIA argues that autonomous-vehicle safety evaluation should connect measured road tests to repeatable simulation and hardware checks, rather than treat them as separate tests. Its Halos SAFE framework reconstructs captured scenarios with Omniverse NuRec, checks simulated results against real-world measurements and vehicle hardware, then uses generated variations to retest software releases. NVIDIA says the loop can produce evidence for safety targets and certification across markets.
A safety case has to connect the road to repeatable tests
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. NVIDIA’s announced Halos SAFE framework is aimed at evaluating safety through a connected process: road testing, simulation, and fleet data feed into one safety loop.
NVIDIA says Halos SAFE provides test suites for NCAP testing worldwide and safety evaluations from L2 driver assistance to L4 robotaxis. The standards shown include US NCAP, Euro NCAP, C-NCAP, JNCAP, and KNCAP. The autonomy levels on screen range from L2 driver assistance and L2++ advanced driver assistance through L3 conditional automation to L4 robotaxis, alongside the UN-R labels DCAS, ALKS, and ADS.
The process begins with measured driving. NVIDIA captures a sample of scenarios on a test track and measures what happened; those measurements become the “ground truth” for the scenario. Omniverse NuRec reconstructs the scene, allowing the same test to be run in simulation. NVIDIA compares the vehicle’s simulated response with the track measurements. When the simulation results match the real-world scenario, the scenario can be reused to test new software releases.
A displayed Vienna example shows how a test can include sensor-level evaluation. Front- and right-camera views of a city street sit beside a top-down visualization of a pedestrian and a safety zone. The dashboard identifies the pedestrian as detected and not in the safety zone.
Simulation is checked against the complete system
Halos SAFE also tests whether improvements seen in simulation carry through to hardware. NVIDIA says this includes DRIVE AGX Thor in a hardware-in-the-loop simulator. The displayed interface combines a simulated driving environment with plots for GPS speed and IMU acceleration.
A separate test puts virtual hazards in front of a real car. The vehicle receives simulated sensor inputs in real time, creating obstacles and edge cases that are not physically present. The car can perceive and react to them, allowing dangerous situations to be tested without placing those hazards on the track.
These checks address different parts of the evaluation: hardware-in-the-loop testing checks whether simulation improvements carry through to hardware, while virtual sensor inputs let a real car respond to hazards that exist only in simulation. NVIDIA says each result is measured against a safety target, so evaluators can identify what passed, what failed, and why. The dashboard shown includes crash-avoidance and corridor checks, measurement data, scores and KPIs, and a scenario viewer.
Each software release returns to the test loop
Findings feed back into development, and the next software version is tested again. NVIDIA Cosmos adds offline variations, generating thousands of versions across lighting, weather, road surfaces, and traffic. The accompanying visual shows simulated drives under varied conditions.
The resulting loop links captured drives, reconstructed scenarios, simulation, hardware checks, and further software testing. NVIDIA presents Halos SAFE as continuous safety validation and as a way to provide evidence for certification across markets. Its stated aim is for each new version to be proven safe and deployed with confidence around the world.