NVIDIA unveiled Halos SAFE, a Safety Assurance Framework for Evaluation that links simulation, software/hardware-in-the-loop testing, on-road testing, and fleet monitoring, developed with partners including JLR, Mercedes-Benz, and Volvo Cars.

At GTC Berlin, NVIDIA announced that European automakers and mobility providers are showcasing autonomous-mobility use cases on the NVIDIA Hyperion platform, signaling potential for level 3 and level 4 vehicles and future robotaxi fleets.

European partners such as Bolt, Lucid, Mercedes-Benz, Volkswagen Commercial Vehicles, Einride, and Volvo Cars are demonstrating Hyperion-based autonomous-mobility use cases, with Volvo Cars adopting Hyperion and the Alpamayo Stack for safe automated driving.

Madrid’s upcoming Lucid Midsize platform, Cosmos, was discussed, though the exact Lucid model for Madrid was not disclosed; if Cosmos isn’t ready, Madrid could utilize existing Gravity or Air vehicles.

Cosmos Data Factory provides a blueprint for AV dataset generation using synthetic data, while Alpamayo Agentic Factory (early 2027) will connect data prep, post-training, and simulation evaluation through AI agents.

Total robotaxi commitments from Uber/Lucid/Nuro and Bolt total at least 60,000 vehicles globally, including 35,000 Gravity for Uber/Nuro and 25,000 for Bolt.

Bolt is launching a European robotaxi program using Lucid vehicles equipped with NVIDIA Alpamayo Stack and Hyperion, with Madrid as the initial operations site in 2027, pending regulatory approval, and a goal to prepare at least 25,000 Level 4-ready Lucid vehicles for Europe, aiming toward 100,000 autonomous vehicles by 2035.

Madrid will begin with data collection to gather real-world driving data for software development and regulatory alignment, with Germany and France named as next markets.

Bolt will lead autonomous ride-hailing development in Europe, adapting software for European roads, coordinating with authorities, starting in Madrid with potential expansion to Germany and France.

NVIDIA’s Cosmos Data Factory, in collaboration with Uber, is used to curate and process AV data; Porsche leads a data flywheel across Volkswagen Group using Cosmos Data Factory and Omniverse NuRec on Microsoft Azure to convert fleet data into training datasets and simulations, with aims to reduce physical testing by up to 90% and speed development by up to 85%.

Porsche drives a data-flywheel across Volkswagen Group brands utilizing Cosmos Data Factory and Omniverse NuRec on Microsoft Azure to generate training data and simulations, targeting substantial reductions in physical test drives and development cycles.

New developer models and tools were announced, including Alpamayo 2 Edge for 10 Hz in-vehicle scene understanding, AlpaJudge for evaluating driving explanations, Foundation4D labels for lidar-free perception, Cosmos 3 Nano variants for synthetic data generation, and the Alpamayo Agentic Factory for streamlined data prep and simulation evaluation.