Foretellix and Voxel51 Partner to Bring Scalable 3D Neural Reconstruction to AI-Powered AV Development

Joint solution enables end-to-end data curation, audit, reconstruction, and scenario variation for AI-powered autonomy

TEL AVIV, Israel and ANN ARBOR, Mich., Dec. 9, 2025 /PRNewswire/ -- Foretellix, the leading Physical AI toolchain provider for safe autonomous vehicles, and Voxel51, the most powerful visual AI data platform, today announced a joint solution that transforms real-world drive logs into high-fidelity 3D scenes and neural reconstructions for training, testing, and validating autonomous vehicle (AV) systems.

As AV developers increasingly adopt AI-driven, end-to-end architectures (AI AV stacks), data quantity and diversity have become a central bottleneck. Real-world driving data is costly and rarely captures the unique or complex events needed to train and validate modern AV stacks at scale. By using neural reconstruction and synthetic sensor data generation to recreate the real-world drives and introduce controlled variations of the drives in simulation, the Foretellix Physical AI toolchain is able to fill the gaps in the real-world driving data. Companies are also realizing that flawed reconstructions or synthetic data generation from poor-quality input data result in weeks of wasted engineering time and millions of compute dollars. Voxel51's Physical AI Workbench ensures that every simulation starts with accurate, high-quality data by transforming multimodal sensor data into high-fidelity simulation-ready datasets. Together, Foretellix and Voxel51 enable teams to build richer, more diverse driving datasets that are grounded in real-world conditions to scale AV development, training, testing, and validation.

Foretellix and Voxel51 Integration Workflow

Foretellix's open ecosystem allows for the Foretify Physical AI toolchain to integrate external technologies such as Voxel51's FiftyOne Physical AI Workbench, which creates a production-grade end-to-end workflow for AV stack testing and validation:

• The pipeline begins with Foretellix's Foretify toolchain ingesting real-world drive logs to evaluate and identify gaps in operational design domain (ODD) coverage.
• Using scenario-driven data curation, Foretellix identifies the most relevant snippets from large drive-log datasets to fill these coverage gaps.
• Then, the real-world drive logs flow through Voxel51's data curation and audit checks to automatically detect pose calibrations, sensor misalignments, coordinate conventions, and poor-quality labels - critical checks for high-quality 3D reconstructions and simulations.
• Voxel51 further enriches the curated data by adding valuable structure and context through embeddings, scene understanding, and visual inspection, preparing it for 3D reconstruction with NVIDIA Omniverse NuRec 3D Gaussian splatting (3DGS)-based reconstruction technology.
• Foretellix then generates controlled scenario variations on top of these reconstructions and produces synthetic sensor data that can be used for training, testing, and validation purposes - including closed-loop sensor-level simulation.
• These datasets are then analyzed within FiftyOne's inspection and visualization environment for both reconstructed and synthetic datasets before downstream model evaluation.
• To complete the cycle, Foretellix's Foretify toolchain analyzes the new and updated drive log dataset to evaluate the ODD coverage and to ensure that the new scenario variations have successfully resolved the originally identified gaps.

"Safety is the foundation that Physical AI depends on. As AV stacks shift toward end-to-end AI, developers need platforms that can generate and manage vast, diverse data at scale. Foretify delivers that foundation with the depth, power, and automation required for the next era of AI powered autonomous systems. Through our work with Voxel51, we unite real-world grounding with controllable scenario variation in a single workflow that empowers teams to build stronger, smarter, and safer AI-based AV systems." said Ziv Binyamini, CEO and Co-Founder of Foretellix.

"We're excited to work with Foretellix to bring high-fidelity reconstructions and synthetic data generation to AV developers," said Brian Moore, Co-Founder and CEO of Voxel51. "As the volume of Physical AI data continues to explode, data quality has become mission-critical to building reliable systems. Poor quality or incomplete data drains resources and poses serious reliability risks in safety-critical AV applications. Together, we're helping teams build AI-powered autonomy systems with greater realism, efficiency, and confidence."

About Voxel51

Voxel51 is the leading computer vision and visual AI data platform that enables the development of safe and reliable visual AI systems. The company's flagship product, FiftyOne, combines open source flexibility with enterprise-grade capabilities to enable organizations to visualize, curate, annotate, and understand their visual data and models. Trusted by millions of developers and thousands of enterprises—including LG Electronics, Microsoft, Berkshire Grey, and Precision Planting—FiftyOne empowers organizations to maximize AI performance with better data.

Learn more at voxel51.com. See us live at TechAD (Dec 8-9, 2025), Booth #21, and at CES 2026 West Hall – Booth #4767.

About Foretellix

Foretellix is the leading enabler of safe autonomous vehicles. Its Physical AI toolchain accelerates the training, validation and safety evaluation of autonomous vehicles, making AI safe for the physical world. Foretellix empowers OEMs and AV developers to build next-generation AV technology with measurable safety, efficiency, and scale. With offices in the US, Europe, and Asia, Foretellix is driving the shift to intelligent, AI-powered autonomy.

For more information, visit foretellix.com or see a live demo at CES 2026 West Hall – Booth #4767.

Logo - https://mma.prnewswire.com/media/1977129/Foretellix_Logo.jpg

Press Contacts: 

Foretellix
George Giles
[email protected]

SOURCE Foretellix