Foretellix has partnered with Voxel51 to link their technologies into a single workflow that converts real-world drive data into detailed 3D scenes and neural reconstructions for autonomous vehicle development.
The combined setup is designed for training, testing and validating end-to-end autonomous driving systems, with a focus on overcoming limits in the volume and diversity of data available to AI-based architectures.
Go deeper with GlobalData
Discover B2B Marketing That Performs
Combine business intelligence and editorial excellence to reach engaged professionals across 36 leading media platforms.
Voxel51 CEO and co-founder Brian Moore said: “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.”
The process starts with Foretellix’s Foretify platform, which takes in real-world drive logs and assesses them for gaps in operational design domain (ODD) coverage.
Foretellix then applies scenario-led data selection to pick out relevant sections of the logs that can help address those gaps.
These chosen segments are then processed by Voxel51’s data curation and audit tools.
At this stage, the logs are checked for issues such as pose calibration errors, sensor misalignment, inconsistencies in coordinate systems and low-quality labelling.
Voxel51 further enhances the filtered data using embeddings, scene interpretation and visual review, preparing it for 3D reconstruction.
The partners’ workflow employs Nvidia Omniverse NuRec technology, which uses 3D Gaussian splatting to build high-fidelity scene reconstructions.
Foretellix subsequently applies controlled scenario variations to the reconstructed scenes and generates synthetic sensor outputs.
According to the companies, this enables training, testing and validation workflows, including closed-loop simulation at the sensor level, using both reconstructed and synthetic data.
Foretellix CEO and co-founder Ziv Binyamini said: “As AV stacks shift toward end-to-end AI, developers need platforms that can generate and manage vast, diverse data at scale.
“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.”
