Voxel51 launches 3.0 platform to streamline physical AI data workflows

Voxel51 has launched Voxel51 3.0, an end-to-end multimodal data platform designed to streamline data workflows for teams developing models using sensor data from robots, vehicles, and factories. The update introduces Voxel51 Agent, a data provenance layer, and free self-service access. The platform now supports multiple modalities, including MCAP, images, video, 3D point clouds, DICOM, NIfTI, geospatial data, and audio. Key features include Voxel51 Search, a persistent hybrid index that automatically indexes datasets and computes embeddings, and Vector retrieval using LanceDB, which separates storage from compute. Voxel51 3.0 also adds four provenance layers and is model-agnostic across embeddings, auto-labeling, inference, and evaluation. The release highlights findings from Voxel51’s 2026 State of Physical AI survey, noting that 58% of 'exceptional' teams spend over half their project time on data work, compared to 21% of struggling teams. The platform’s capabilities include semantic retrieval, which finds visually or semantically similar moments without shared labels, and lexical and structured queries for exact terms, timestamps, and metadata. Voxel51 Agent enables repeatable workflows by chaining search, curation, annotation, and evaluation. The platform also allows teams to build custom plugins and integrate with governance controls. Voxel51 cites Encord’s sensor-stream support from its Series C release, with prices verified on October 17, 2026.