Voxel51 Taps NVIDIA Technology to Power Industry’s First Physical AI Data Engine for Simulation-ready Workflows

First standardized data engine integrated with NVIDIA Omniverse and NVIDIA Cosmos to standardize and accelerate the creation of high-quality simulation-ready Physical AI datasets

SAN FRANCISCO, Nov. 05, 2025 (GLOBE NEWSWIRE) -- Voxel51, the most powerful visual AI data platform, announced today the general availability of the FiftyOne Physical AI Workbench, the industry’s first standardized data engine for processing and validating real-world data for simulation and model training. As simulation becomes central to developing safe and scalable physical AI systems, from autonomous vehicles to humanoid and industrial robots, variable data quality has created a bottleneck. Flawed simulations based on poor or incomplete data not only undermine system performance but also waste vast amounts of compute and engineering resources. The FiftyOne Physical AI Workbench eliminates wasted costs by ensuring data quality upfront, streamlining the entire data pipeline from ingestion and validation to reconstruction and synthetic data generation.

Building AVs that navigate complex traffic scenarios or robots that safely maneuver manufacturing environments requires repeated testing in virtual worlds before being ready for the physical world. By integrating NVIDIA Omniverse and Cosmos, the FiftyOne Physical AI Workbench enables developers to curate large datasets, prepare data for reconstruction, and amplify data variation. Using Cosmos Dataset Search (CDS), developers can quickly find relevant and interesting scenarios within the massive pool of recorded data. This data can then be reconstructed and scaled using NVIDIA Omniverse NuRec neural reconstruction libraries and the NVIDIA Cosmos Transfer world foundation model; however, developers must be able to ensure data quality first.

Companies are realizing that slight misalignments or miscalibrations, even just microseconds off, in multi-sensor perception data streams like camera-to-LiDAR, radar, or geolocation, could render sophisticated simulations inaccurate and unusable. The outcome is costly: weeks of wasted engineering time and millions of dollars in compute resources lost to errors that could have been caught and fixed earlier. In fact, according to Gartner, “poor data quality costs organizations an average $12.9 million” every year. Globally, these costs could compound to billions annually.

The FiftyOne Physical AI Workbench ensures that every simulation starts with accurate, high-quality data. The workbench detects input data errors such as sensor miscalibration, synchronization issues, and poor-quality labels — all of which can lead to flawed simulations and downstream model failures. Teams can easily visualize, search, and enhance validated data to seamlessly generate realistic 3D recreations and synthetic variations, including hazardous conditions or rare-edge cases in simulation.

“Our customers tell us that over 50% of simulations fail due to bad input data. Teams are burning millions of dollars on compute, only to realize that their simulation results are unusable,” said Brian Moore, co-founder and CEO of Voxel51. “Data is the foundation for successful Physical AI. Poor quality and incomplete data not only drain resources but every failed simulation adds weeks to development. The workbench solves this by giving engineers a standardized platform to bring diverse, high-fidelity datasets into simulation pipelines.”

FiftyOne Physical AI Workbench offers a fully integrated pipeline that unifies multi-sensor data in a single workspace:

• Audit and Validate: Performs automatic calibration, sensor alignment, metadata consistency checks, and human-readable QA reports.
• Enrich: Adds structure and context through embeddings, auto-labeling, scene understanding, and visual 3D inspection and search.
• Generate: Normalizes data for reconstruction and augmentation, integrating directly with leading tools such as NVIDIA Omniverse NuRec, NVIDIA Cosmos Dataset Search, and Cosmos world foundation models.
  
  

“By combining NVIDIA Omniverse, NVIDIA Cosmos, and Voxel51’s Physical AI Workbench, we’re advancing the data infrastructure needed for physical AI,” said Norm Marks, Vice President of Automotive at NVIDIA. “This unified data engine transforms complex data into simulation-ready intelligence, key to efficient AV development.”

FiftyOne Physical AI Workbench is available immediately for enterprise customers. For more information, visit https://voxel51.com/physical-ai/.

About Voxel51
Voxel51 provides an AI development platform that helps organizations unlock the full potential of their visual data. Our open source and commercial software, FiftyOne, is trusted by millions of AI builders and thousands of leading enterprises––including LG Electronics, Microsoft, Berkshire Grey, and Precision Planting––for its intuitive and powerful approach to developing high-performing AI systems. By leveraging innovative techniques to explore, refine, and enhance large-scale datasets and models, Voxel51 enables organizations to deliver accurate and reliable AI models that drive real-world impact. Learn more at Voxel51.com.

Media Contact
Stephanie Schlegel
Offleash PR for Voxel51
voxel51@offleashpr.com

Legal Disclaimer:

EIN Presswire provides this news content "as is" without warranty of any kind. We do not accept any responsibility or liability for the accuracy, content, images, videos, licenses, completeness, legality, or reliability of the information contained in this article. If you have any complaints or copyright issues related to this article, kindly contact the author above.