Rerun

What is Rerun?

Rerun is a comprehensive data layer designed specifically for Physical AI, enabling developers to build, understand, and improve their data loops. It handles multi-rate, multimodal data from initial recording to massive scale, providing a flexible framework that adapts to your workflow rather than a rigid platform. With Rerun, you can review datasets, debug detailed issues, and extend functionality with custom views and tools at every pipeline stage.

Key features include full dataframe and SQL query capabilities over any robotics data, allowing you to analyze and manipulate data seamlessly. You can extend datasets with post-processing or annotations as easily as adding columns to a table. Data is stored as column-chunks in rrd files, supporting efficient storage and retrieval. Rerun also offers an open-source SDK for logging data directly or converting from other formats.

Rerun Hub is the production backend that serves as a data catalog and storage engine. It enables cataloging, byte-range indexing, and retrieval, turning your object stores into a queryable, streamable foundation. You can run transforms on the edge or close to the data source. Query capabilities extend across your catalog, down into columns, time ranges, and values within recordings. Transforms allow adding derived columns and evolving schemas without breaking history, with the SDK handling execution while Hub organizes derived and raw data together.

For training, Rerun lets you express a dataset mix as a query and stream it directly to GPUs. The dataloader is column-aware and video-codec-aware, enabling direct training on recordings. Sharing is simplified with a single viewer that allows teams to explore, annotate, and trace failures back to the originating data.

Rerun is suitable for robotics, autonomous vehicles, and any physical AI application requiring robust data management. Technical details include support for multi-rate data, column-chunk storage, and integration with object stores. The recent 0.32 SDK release unifies visualization, querying, transformation, and training, making Rerun a complete data layer for physical data.

Use Cases

Key Features

multi rate multimodal data full dataframe queries SQL over robotics data extend datasets with annotations column chunk storage open source SDK Rerun Hub catalog byte range indexing edge transforms custom views and tools

Who is it for?

robotics developers, AI engineers, data scientists, simulation teams, perception engineers, autonomy researchers, ML pipeline builders

Similar Tools