rerun-io/rerun is a tool that visualizes and queries multimodal robot data and streams it so it can be used for training. It is written in Rust and released under the Apache-2.0 license, and its current latest release is 0.39.0. It connects the flow of collecting data, inspecting it with your own eyes, and then feeding it into training within a single tool.
The README describes Rerun as a data layer for physical AI. It records multi-rate, multimodal data coming from multiple sources into shared columnar storage, where it can be queried and visualized and streamed directly into training. Multi-rate means handling signals with differing periods as a single flow. The built-in viewer synchronizes and displays images, point clouds, transforms, time series, joint states, and video in real time, and the data can be queried with dataframes or SQL. It provides Python, Rust, and C++ SDKs. The repository, created on April 8, 2022, has 11614 GitHub stars.
However, the provided research record contains no performance benchmarks or real adoption cases, and neither the changelog for 0.39.0 nor its public release date is given. The momentum score was measured low at 2.1, and the novelty, technology, and quality scores are not recorded. Therefore, from this record alone, it is impossible to judge how much load appears as data volume grows, or how far it is actually used in practice.