Voxel51 has officially released Voxel51 3.0, updating its data platform to unify search, curation, annotation, and evaluation for teams building physical AI applications. The update expands support beyond visual data to include multimodal sensor inputs such as 3D point clouds, MCAP, DICOM, meshes, and audio, catering directly to robotics, autonomous vehicle, and industrial engineering teams.

The platform introduces Voxel51 Search, a persistent hybrid index powered by LanceDB that automatically computes embeddings and handles vector, lexical, and structured queries. Additionally, Voxel51 Agent enables users to automate complex data workflows and generate custom plugins while maintaining organizational governance and access controls.

According to company research cited in the announcement, 58 percent of high-performing physical AI development teams spend over half their project duration on data management. Voxel51 3.0 aims to streamline this overhead by offering a model-agnostic, self-service platform where teams can integrate foundation models or plug in proprietary domain models.

Why it matters

  • Multimodal data tooling is becoming critical infrastructure for robotics and physical AI deployment across industrial sectors.

  • Hybrid vector-lexical indexing reduces data engineering overhead, allowing founders and operators to iterate faster on domain-specific datasets.

Source: marktechpost.com