Volkswagen Group Builds Shared AV Data Flywheel With NVIDIA, Microsoft, Voxel51
Porsche, Volkswagen Group brands and CARIAD are jointly building a shared data flywheel for autonomous vehicle development, together with NVIDIA, Microsoft and Voxel51. The collaboration is expected to reduce repeated physical test drives by up to 90% and shorten AI regression test cycles by up to 85%, according to the announcement.
- Porsche, Volkswagen Group brands and CARIAD are building a shared AV data flywheel with NVIDIA, Microsoft and Voxel51, announced at NVIDIA GTC Berlin.
- The companies expect up to 90% fewer repeated physical test drives and up to 85% shorter AI regression test cycles.
- Voxel51 audits raw drives with roughly 80 checks before NVIDIA Omniverse NuRec reconstruction.
- Workloads run on Microsoft Azure within Volkswagen Group's existing security and governance model.
A reference architecture built on fleet data
The companies announced the project at NVIDIA GTC Berlin. Porsche leads the initial implementation and the ecosystem integration. According to the announcement, the Volkswagen Group has a fleet that generates data equivalent to 11,000 average drivers per day, and it is using that data to build a physical AI stack linking fleet data to simulation, training and evaluation.
Each partner has a defined role:
- Volkswagen: contributes its connected fleet, real-world driving data, vehicle platform and system requirements.
- Voxel51: provides the data layer, turning raw fleet data into training and simulation inputs and linking model failures back to the scenarios and data needed for the next iteration.
- NVIDIA: supplies data retrieval, generation and simulation technologies, including Cosmos Data Factory and Omniverse NuRec libraries for reconstructing real-world drives.
- Microsoft Azure: provides cloud and compute infrastructure that Volkswagen already uses.
The companies describe the result as a reference architecture for automakers building the next generation of advanced driving systems.
Voxel51 audits drives before reconstruction
Virtual validation of driving-software releases is one of the first applications. The team renders reconstructed camera inputs from real drives and compares simulated outputs with real-world measurements. The stated aim is to make testing repeatable, shorten feedback cycles and reduce dependence on physical test drives.
The release stresses that reconstruction is sensitive to input errors. A camera calibration that is off by a few degrees or a LiDAR clock offset will not stop a reconstruction job, but it can produce a scene that scores fine and looks wrong after hours of GPU time, and every downstream simulation inherits the error. To prevent this, Voxel51 audits each raw recording across roughly 80 checks covering sensor integrity, timing, calibration, trajectories, actors and coordinate frames.
The audit reads MCAP files directly, and the fastest checks complete in a fraction of a second per drive. Each check records whether it passed, failed or could not run, so missing inputs are not mistaken for valid data.
Engineers can launch and track NVIDIA Omniverse NuRec reconstructions and data from NVIDIA Cosmos runs from within Voxel51. Each synthetic output is automatically checked for quality and stays linked to the source clip, generation recipe and model that produced it. Quality is evaluated across several dimensions rather than a single score, with image quality measured by camera and object class, and separate checks for vehicles and pedestrians. Each run then enters a review queue, where engineers can approve it for simulation, send it back for retraining or reject it.
Why the companies see fleet data as the advantage
Voxel51 also powers workflows including multimodal search and retrieval, automated labeling, scenario-based curation, dataset balancing, training data selection and model validation. Each stage runs as a job on Microsoft Azure, where the Volkswagen Group manages its fleet data, inside its existing security, identity and regional governance model.
Voxel51 CEO Brian Moore said: "The key to really industry-leading performance is to harness the data that your company has access to." Rishi Dhall, vice president of automotive at NVIDIA, said Volkswagen Group can use the platform and NVIDIA technology to "turn fleet data into high-fidelity simulation scenarios to accelerate the development and validation of next-generation driving systems."
A common data model lets driving data and scenarios created in one project carry forward across vehicle programs, software releases and brands. According to the announcement, every new drive expands the scenario base, and reconstructed scenes can be reused in later development and validation cycles.
What it means for automakers
The announcement frames the opportunity for automakers as defined not only by how much proprietary data they hold, but by how quickly they can turn road events into the next training, simulation and validation workload. The 90% and 85% reductions are expected outcomes stated by the companies, not measured results. For other automakers, the project is presented as a template for connecting fleet data, simulation and cloud infrastructure. Voxel51 says its platform is already used by teams at Berkshire Grey, Porsche, Volkswagen Group, Ford and Precision Planting.
